Access to this full-text is provided by IOP Publishing.
Content available from Physics in Medicine & Biology
This content is subject to copyright. Terms and conditions apply.
Phys. Med. Biol. 70 (2025) 015019 https://doi.org/10.1088/1361-6560/ad9db0
Physics in Medicine & Biology
OPEN ACCESS
RECEIVED
25 September 2024
REVISED
24 November 2024
ACC EPT ED FOR PUB LICATI ON
11 December 2024
PUBLISHED
6 January 2025
Original Content from
this work may be used
under the terms of the
Creative Commons
Attribution 4.0 licence.
Any further distribution
of this work must
maintain attribution to
the author(s) and the title
of the work, journal
citation and DOI.
PAPER
Design, construction and validation of a magnetic particle
imaging (MPI) system for human brain imaging
Eli Mattingly1,2,4,∗, Monika ´
Sliwiak1, Erica Mason1,4, Jorge Chacon-Caldera1,4, Alex Barksdale1,3,
Frauke H Niebel1,6,7, Konstantin Herb5, Matthias Graeser6,7and Lawrence L Wald1,2,4
1Dept. of Radiology, A.A. Martinos Center for Biomedical Imaging, Massachusetts General Hospital, Charlestown, MA, United States
of America
2Harvard-MIT Division of Health Sciences & Technology, Cambridge, MA, United States of America
3Department of Electrical Engineering & Computer Science, Massachusetts Institute of Technology, Cambridge, MA, United States of
America
4Harvard Medical School, Boston, MA, United States of America
5Department of Physics, ETH Zurich, Zurich, Switzerland
6Fraunhofer Research Institution for Individualized and Cell-Based Medical Engineering, Fraunhofer IMTE, Lübeck, Germany
7Institute of Medical Engineering, University of Lübeck, Lübeck, Germany
∗Author to whom any correspondence should be addressed.
E-mail: Eli.Mattingly.phd@gmail.com
Keywords: magnetic particle imaging, mpi, superparamagnetic iron oxide nanoparticles, cerebral blood volume,
functional neuroimaging, SPION, brain imaging
Supplementary material for this article is available online
Abstract
Objective. Magnetic particle imaging (MPI) was introduced in 2005 as a promising, tracer-based
medical imaging modality with the potential for high sensitivity and spatial resolution. Since then,
numerous preclinical devices have been built but only a few human-scale devices, none of which
targeted functional neuroimaging. In this work, we probe the challenges of scaling the technology
to meet the needs of human functional neuroimaging with sufficient sensitivity for detecting the
hemodynamic changes following brain activation with a spatio-temporal resolution comparable to
current functional magnetic resonance imaging approaches. Approach. We built a human
brain-scale MPI system using a mechanically-rotated, permanent-magnet-based field-free line
(FFL) (1.1Tm−1) with a water-cooled, 26 kHz drive coil producing a field of up to 7 mTpeak, and
receive coil that can fit over a human head. Images are acquired continuously at a temporal
resolution of 5 s/image, controlled by in-house LabView-based acquisition software with online
reconstruction. We used a dilution series to quantify the detection limit, a series of parallel-line
phantoms to assess the spatial resolution, and a large ‘G’ shaped phantom to demonstrate the
human-scale field of view (FOV). Main results. The imager has a sensitivity of about 1 µgFe over a
2D imaging FOV of 181 mm diameter(132 pixels) in a 5 s image. Depending on the image
reconstruction used, the spatial resolution defined by 50% contrast between adjacent lines was
5–7 mm. Significance. This proof-of-concept system demonstrates a pathway for human MPI
functional neuroimaging with the potential for an order of magnitude increase of sensitivity
compared to the other human hemodynamic imaging methods. It demonstrates the successful
transition of the FFL based MPI architecture from the rodent to human scale and identifies areas
which could benefit from further work.
1. Introduction
There are various in vivo neuroimaging modalities for applications ranging from anatomical structure
studies to measurements of functional tissue properties. Each technique has a unique combination of
biological contrast specificity, tissue-penetration capability, signal sensitivity, spatiotemporal resolution, as
well as its injected agents’ shelf-life and toxicity profile. magnetic particle imaging (MPI) was introduced in
© 2025 The Author(s). Published on behalf of Institute of Physics and Engineering in Medicine by IOP Publishing Ltd
Phys. Med. Biol. 70 (2025) 015019 E Mattingly et al
2005 as a noninvasive imaging modality to map the distribution of injected superParamagnetic iron oxide
nanoparticles (SPIONs) by exploiting their nonlinear response to an applied oscillating magnetic field.
Localization is achieved by superimposing gradient fields that result in signals only arising from a small
field-free region that is swept over the imaging space (Gleich and Weizenecker 2005). MRI detects nuclear
spin magnetism directly and injected magnetic contrast agents such as gadolinium compounds or SPIONs
only indirectly through their relaxation effect on the surrounding water’s nuclear spin resonance. Unlike
MRI, MPI detects the SPION’s magnetization directly. MPI and MRI share the use of applied magnetic fields
and the detection of time-varying magnetic moments through Faraday induction. However, as a tracer
detection method, it is more appropriately classified with other tracer mapping modalities like single photon
emission computed tomography and positron emission tomography.
The SPION detected by MPI will not cross the healthy blood-brain barrier, and therefore an MPI brain
image reflects the cerebral blood volume (CBV). This makes it an attractive technology for imaging the
cerebral vasculature, as well as blood volume, flow and perfusion (Wu et al 2019, Mason et al 2023). The
signal level is proportional to the CBV and is unobscured by non-blood pool signals which contribute to CT
or MR images. The long circulation times of coated SPIONs (Khandhar et al 2017) provides the ability to
regionally monitor the CBV for changes. MPI has been used in this way to detect murine gut bleeds (Yu et al
2017), perfusion in murine stroke (Ludewig et al 2017), intracranial hemorrhage (Szwargulski et al 2020),
and traumatic brain injury in preclinical studies (Orendorff et al 2017). More recently, MPI has been assessed
for mapping CBV changes following hypercapnia in rats (Mason et al 2023).
MPI detects and spatially localizes the magnetization changes in an injected SPION tracer in response to
an applied ‘drive’ field, which is typically (though not necessarily (Tay et al 2019a)) harmonically pure. Since
the SPION magnetization response to the applied drive field is nonlinear (modeled by the Langevin function
in the relaxation-free case (Shliomis 1972, Kluth 2018)), the magnetization signal detected by the receive coil
via Faraday induction contains harmonics of the drive waveform. Application of a field gradient, either in the
form of a field-free-point or field-free line (FFL) (Weizenecker et al 2008, Knopp et al 2010a, Greiner et al
2022) saturates the particles outside of the zero-crossing region, reducing their Faraday response and thereby
localizing the signal to the field-free region. The field-free region is then swept in space to provide the map of
SPION distribution, which can be reconstructed with either a forward model inversion of a measured
(Weizenecker et al 2009) or simulated linear model of the signal (Rahmer et al 2009, Knopp et al 2010b), or
by mapping the signal to the field-free region’s trajectory in space (x-space MPI) (Goodwill and Conolly
2010,2011).
For successful functional neuroimaging of the CBV response to brain activation in humans, the device
should be capable of imaging a human head with 6 mm spatial resolution every 5 s over the course of 30 min.
This minimum spatial resolution reflects that used in many useful functional magnetic resonance imaging
(fMRI) studies which typically smooth to a 4–12 mm spatial resolution (Huettel et al 2014). Other fMRI
studies have also well-characterized the timecourse of CBV changes (Mandeville et al 1998, Mason et al
2023), and show that the hemodynamic response reaches 90% of its peak in about 5s (DeYoe et al 1994),
therefore this or better temporal resolution is needed to capture the dynamic changes. The functional MPI
scanner needs to be able to stably image for at least 30 min to perform studies similar to a typical fMRI study.
While these specifications offer no advantage over fMRI, which also offers a wealth of useful anatomical and
diffusion contrasts, theoretical analysis (Mason et al 2017) and preliminary fMPI studies in rats (Mason et al
2023) suggest that the sensitivity of fMPI can potentially exceed that of fMRI, due to its detection of the
much larger SPION magnetization (compared to the hydrogen nuclear magnetic moment) and the lack of
confounding signals from the non-blood pools (which contribute physiological noise to fMRI (Triantafyllou
et al 2006, Krüger et al 2010)).
Although no human MPI has been performed, there are currently a few systems under development for
human-scale MPI. Each has been tailored for a separate application and thus would be poorly suited for
fMPI. Examples include a device tailored for the imaging of stroke perfusion deficits in a portable setting
which employed low power and achieved portability at a cost of spatial resolution (Graeser et al 2019,
Thieben et al 2024). Others have built systems for interventional radiology of limbs, an application which
does not require high sensitivity or long-term stability (Vogel et al 2023). Other systems have been presented
at early stages of development (Top et al 2017, Mason et al 2024, Nomura et al 2024, Yoshida et al 2024) but
very little experience exists with human-scale MPI scanner hardware, especially for functional neuroimaging
applications.
In this work, we introduce a mechanically rotating FFL-based Magnetic Particle Imager designed for
human brain imaging. We describe the design of each subsystem, including the overall mechanical gantry
which rotates the permanent-magnet-generated FFL and high-power electromagnet shift coils, a 26.3 kHz
drive coil which generates the SPION response and the drive chain filter, as well as the receive coil and low
2
Phys. Med. Biol. 70 (2025) 015019 E Mattingly et al
Figure 1. Top left: Digital rendering of a cross-section of the imager with major dimensions indicated. The drive coil’s inner
diameter is 27 cm and the major and minor axes of the elliptical receive coil are 23 cm and 16cm (not shown). Top right:
photograph of the imager from a similar angle. Bottom: Schematic of the main subsystems and an illustration of the system. The
primed coordinated directions (i.e. x’, y’) indicate the rotating coordinate frame, where θis the gantry rotation angle.
noise electronics. We characterize its spatial resolution, sensitivity, field of view (FOV) and temporal stability
in phantom imaging, including phantoms at the human brain scale.
2. Methods
2.1. MPI scanner and acquisition scheme
2.1.1. Architecture of the human scale MPI scanner
The system architecture refines our previously presented rodent-scale imager (Mattingly et al 2022a) and
scales it to a geometry suitable for human head imaging. A preliminary version of the scanner has been
presented in abstract form (Mattingly et al 2024b). The scanner is based on a FFL architecture (Weizenecker
et al 2008), with the FFL gradient generated by a pair of mechanically rotated NdFeB permanent magnets. An
electromagnetic coil pair shifts the zero-crossing of the FFL’s gradient across the human head (±10cm) to
form a projection at each rotation angle. A single drive coil produces an oscillating field at 26.3kHz to drive
the SPIONs in and out of (partial) saturation. The dM/dtresponse is recorded as the EMF induced in a
single first-order gradiometer receive coil. In this receive coil, the head occupies the windings of one of the
counterwound pair. Figure 1shows an illustration of a head in the coil, a system schematic, a cross-sectional
drawing, and a scanner photograph.
2.1.1.1. Mechanical rotating gantry:
The rotating gantry serves to move the FFL through the radial projection angles needed for the 2D image.
The roughly 1500 kg circular rotating assembly includes the FFL’s permanent magnets and the shift coils. The
assembly rotates on the two 1.9 m diameter outer rings formed from 50.8 mm square solid aluminum bars.
These were rolled into semicircles, and assembled with dielectric breaks along the x’–zplane (the primed
coordinate directions are the rotating coordinate frame as in figure 2) to reduce eddy currents from the
time-varying shift fields. They roll on 8 polyurethane rollers supported by a similarly welded aluminum
stand. A frame extension on the service end of the scanner (+zend) supports the rotating portion of the
electrical slip rings (MT080-P16300-1KV, Moflon Technology, Shenzhen, China) and cooling water’s rotary
union (MEPH200-09-ID80, Moflon Technology, Shenzhen, China) which allows delivery of current and
3
Phys. Med. Biol. 70 (2025) 015019 E Mattingly et al
Figure 2. Top left: Simulation of the permanent magnets showing the location of the FFL. Top right: Photograph of the fully
constructed shift coil (background removed for clarity). Bottom: Dimensions of the shift coils and FFL magnets. The primed
coordinated directions (i.e. x’, y’) indicate the rotating gantry coordinate frame, where the axis of rotation is z.
water to the rotating shift coils. This allows unlimited rotation without having to reverse the gantry to
unwrap cables and water lines. This is essential because of the large inertia of the system and 5 s temporal
resolution target. A fully electrically rotated FFL is possible (Weizenecker et al 2008), but would require
considerably more electrical complexity and power requirements. Optimizing the electromagnetic design
can temper the power requirements, though not entirely (Erbe et al 2011). Alternatively, a rotating Halbach
array could have been used for the generation and rotation (Weber et al 2018, Ergor and Bingolbali 2022),
though these designs are more mechanically complex and may be challenging to adapt to this large scale.
Single-sided MPI systems have also been suggested for human imaging, but lack the depth penetration and
scan volumes necessary (Sattel et al 2009, Pagan et al 2021), and are generally more well-suited for specialized
applications such as breast cancer imaging.
The rotation is powered by a 3 HP electric motor coupled to the gantry by a shaft passing through the
wall of the shielded room (which houses the scanner). The shaft is coupled to a timing belt and pinion
engaging with a large (1.9 m diameter) circumferential gear bolted to the gantry ring. With the motor at its
operating speed of 580 RPM, the 1:10 worm gear on the motor together with the 16:155 gearing on the wheel
rotates the gantry at 6 RPM (one rotation every 10 s) allowing an image time-series to be acquired with a
temporal resolution of 5 s (an image every 180◦). The gantry angular position is tracked with an angular
encoder as well as an optical homing switch. The homing switch allows for images to be triggered at
consistent angular positions, and therefore prevents angle offset errors from adding image instability.
2.1.1.2. FFL magnets:
A gradient above 1 T m−1is needed to achieve the target spatial resolution of 5–6 mm given the
magnetization curve of the commonly used, high performance SPIONs such as Synomag-D (Micromod
Partikeltechnologie GmbH, Rostock, Germany), without deconvolving the point-spread function, which
comes at the cost of sensitivity as the deconvolution operation is ill-posed. Synomag-D particles have a
full-width half maximum magnetization response of about 6 mT for the third harmonic component of the
signal as measured with our in-house magnetic particle spectrometer using the ‘system matrix mode’. This
mode of operation records the amplitude of the harmonic signal with a slowly varying (11 Hz) external bias
field (Mattingly et al 2024a). This would be the through-plane kernel (i.e. kernel in z), the MPS is not
currently configured to measure the in-plane (x’–y’) kernel. The FFL magnets, a custom assembly by BJA
Magnetics, Leominister MA, was formed from two 850 mm long NdFeB rare-earth permanent magnets
arranged in opposition and attached to a steel back-plate. Figure 2shows the configuration of these magnets
as well as the dimensions of the assemblies. The FFL magnets consist of 204 (17 ×3×4 in y’,z, and x’
respectively) NdFeB blocks each measuring 50 ×50 ×56.25 mm3(magnetized along the 56.25 mm
4
Phys. Med. Biol. 70 (2025) 015019 E Mattingly et al
Table 1. Measured and simulated resistance and inductance of the shift coils. The simulation values were performed using Ansys
Maxwell at 7 Hz, and the measurements were done at 100 Hz using the Agilent 4263B LCR Meter.
Inner racetrack (‘A’) Outer racetrack (‘B’)
Measured Simulated Measured Simulated
Resistance (DC, 25 ◦C) 310 mΩ301 mΩ399 mΩ385 mΩ
Inductance 26 mH 29 mH 41 mH 43 mH
Measured Simulated
Mutual inductance (inner and outer) 22.6 mH 25 mH
dimension), resulting in an assembly that is 850 ×150 ×225 mm3. Each of the constituent blocks are coated
in an insulating epoxy to prevent the induction of large eddy currents by the time-varying shift fields. The
steel back plate reduces the reluctance of the effective magnetic circuit thereby increasing the gradient
strength as well as partially containing the fringing fields. The back plate is formed from laminated M6
transformer steel to reduce eddy currents. Both the field and eddy current heating were simulated in Ansys
Maxwell (Ansys Inc. Canonsburg, PA, USA). Further discussions on the magnet designs are found in
reference (Mason 2020). The gradient achieved in the constructed magnet was characterized using a 3-axis
Hall magnetometer (Metrolab Technology SA, Geneva, Switzerland, model Metrolab THM 1178) and a
robotic positioner. The measured FFL gradient strength in zis 0.85 T m−1and 1.13 T m−1in x’.
2.1.1.3. Shift coils and amplifiers:
The shift coils produce a roughly uniform field which ramps slowly (2.7Hz rounded triangular waveform) to
sweep the zero-crossing of the FFL gradient across the imaging volume, generating a 1D projection at each
rotation angle. The shift coil is composed of 4 coils; an inner and outer ‘racetrack’ coil (labeled A and B in
figure 2, respectively) on each side of the head. Each racetrack coil has 7 turns per layer and 30 layers total, all
built from Kapton-wrapped copper hollow conductor wires (6 ×6 mm2square cross-section). Water coolant
flows through the hollow-core wire in its 3mm diameter central axial hole. Breaking the shift coils into an
inner and outer coil doubles the number of parallel water circuits. Coolant flow is increased since each
circuit’s length is reduced by a factor of two. The configuration also requires lower voltage amplifiers
(although 4 of them). The simulated and measured electrical impedances of the shift coil racetrack coils are
in table 1. The impedance was measured with an LCR meter at 100 Hz. To determine the mutual inductance,
we connected an inner and outer racetrack in series such that their fields aid each other, and measured the
total inductance to calculate the mutual inductance (Ltotal =Louter +Linner +2M). The expected heat load for
the 250 Apeak triangle wave needed for a FOV of 19cm is calculated from the measured resistances to be
29.5 kW and needs to be removed by the water system.
Four switch-mode amplifiers (International Electric Company Oy, Helsinki, Finland, model MPS
300–750) power the four shift coils. With a maximum current of 300 Apeak, each amplifier is capable of
driving a racetrack coil at the 250 Apeak and 100% duty cycle needed for continuous 19 cm FOV imaging.
The shift amplifiers are fully isolated from the operating console via optical isolation for both digital
interfaces (e.g. enable lines) and analog signals (input and voltage/current monitors) to isolate the
acquisition and control electronics from potential noise from these switch-mode amplifiers. The digital
signals are isolated with digital optoisolators, and the analog signals are isolated via ISO224 isolation
amplifiers (Texas Instruments, Dallas, TX, USA).
2.1.1.4. Bore tube and shielding:
Figure 3shows the 40 cm diameter copper shielding tube used to isolate the sensitive receive electronics from
noise generated by external sources such as the shift coils (or elsewhere). The tube is stationary
(non-rotating) and is positioned between the rotating shift coils and the drive (and receive) coils. It is
constructed from 17 copper rings, each 1 mm thick and 38 mm wide with electrically isolated gaps between
them to prevent the induction of eddy currents by the time-varying shift fields. Each ring is electrically
connected along a single ‘spine’ to ensure they are all grounded to the same point. Considerable currents
circulate in these rings because this shield is strongly coupled to the drive coils. As a point of reference, the
FEMM 4.2 simulation show that a single representative 38mm wide shield loop must support a current as
high as 237 Apeak for a drive current of 50Apeak in the drive coil. As the rings are in a strong magnetic field
(from the shift, FFL magnets, and drive field) any eddy currents present experience substantial Lorentz
forces. They are mechanically stabilized by wrapping them in epoxy-impregnated fiberglass sheets. Without
this support, the rings would be free to mechanically deform in a manner coherent with the drive field,
5
Phys. Med. Biol. 70 (2025) 015019 E Mattingly et al
Figure 3. Photograph of the shielded bore before and after the fiberglass wrapping. The rings are electrically connected along a
single grounded ‘spine’ formed from a hollow conductor copper wire (4mm OD, 2 mm ID). Besides that one connection point,
the rings are electrically isolated to limit large eddy current loops induced by the shift coils’ dB/dtwhile still effectively shielding
the receive coils from external fields in the Zdirection.
introducing a nonlinearity in the drive field that could generate harmonics. These harmonics would be
picked up by the receive coils as a background signal.
2.1.1.5. Drive coils, amplifier, and filter:
The drive coil was designed to achieve 7 mTpeak amplitude when driven with 60 Apeak of current at 26.3 kHz.
A period of exactly 38 ms is used to ensure each period has an integer number of samples and is digitized in
an identical manner. The drive coil fits within the 40 cm diameter shielding tube, and has a 27 cm inner
diameter to fit the head-sized receive coil.
The drive coil consists of four individual winding ‘modules,’ each consisting of 18 turns of 4mm
outer diameter hollow copper tubing (2mm inner diameter for coolant flow) as seen in figure 4. The spacing
of these four modules was optimized for homogeneity along the center axis, weighting the optimization
function by the receive coil’s sensitivity profile (i.e. the drive field is most uniform where the receive coil
is most sensitive). The presence of the 40cm diameter shielding tube strongly affects the drive coil’s magnetic
field. The coils were simulated and optimized with an in-house developed Biot–Savart solver. This solver
accounts for induced eddy currents in the shield by modeling the system as a matrix of coupled inductors,
then using an equivalent circuit model (Mattingly 2024). The solver was written in the Julia Programming
Language8, and following these simulations, FEMM 4.2 was used to verify the findings. The total inductance
of the drive coil when inside the bore is 643 µH. The inductive reactance of each drive module was nulled with
a series capacitor bank totaling 244 nF and composed of 4 CSM Nano capacitors (Celem, Jerusalem, Israel)
to distribute the voltage drop across the coil. The capacitors were connected using green-laser-welded copper
tabs that are soldered to, to improve the connection linearity and robustness because connection nonlinearity
(e.g. from screw connections) can introduce signal instability (Wilkerson 2010, Aderhold et al 2023).
Despite its higher electrical resistance than a similar-sized Litz wire bundle, the hollow conductor wire
with cooling water was chosen to facilitate cooling. Using analytical models for the viscous resistance of the
water flowing within the hollow conductors (Zigrang and Sylvester 1985) at 40 PSI, we calculate the thermal
resistance for each of the four modules is 60 mK W−1, for a net effective thermal resistance of 15 mK W−1.
With 50 Apeak current (typical operating conditions) the drive coils dissipate about 1.6 kW, yielding an
expected temperature rise of 25 K.
The drive amplifier is an AE Techron 8504 (AE Techron, Elkhart, IN, USA), which is a 4kW switch-mode
amplifier. Because the amplifier has switching noise at multiples of 125kHz, we have a low-pass filter with a
cutoff frequency of about 70 kHz immediately after the amplifier to shunt those frequencies out of the signal.
Following that filter, the power is coupled to the main filter with a ferrite transformer to provide
8available at github.com/EliMattingly22/Biot.jl.
6
Phys. Med. Biol. 70 (2025) 015019 E Mattingly et al
Figure 4. Left: Photograph of the drive coil, shown outside of the bore. Center: FEMM 4.2 axisymmetric simulation of the
magnetic fields produced by the drive coil with 60 Apeak at 26.3 kHz. Right: Photograph of the receive coil with dimensions given
in mm. Note: all three are not at the same scale.
common-mode noise isolation and mitigate grounding issues. The transformer core is a pair of U-shaped
N87 ferrite cores (U 141/78/30, TDK Lambda, Tokyo, Japan) which were glued together with a ∼1 mm gap
between the two halves to form a toroidal shape. The measured magnetizing inductance is 1.55 mH and
turns ratio is 1.87:1 (inductance ratio =1:3.51, where the high-turns side is the filter side). The
primary-referred leakage inductance is approximately 320µH, based on the capacitor which resonated with
that leakage inductance. We were not able to measure any increased distortion in the current due to the
ferrite transformer using Rogowski coil current monitors (Ramboz 1996, Ferkovic and Ilic 2007) located
before and after the transformer.
Following the transformer, the drive filter is similar to the design presented previously (Mattingly et al
2022b). The filter is fully balanced, has a T-type band-pass section, then notch sections for the 2nd and 3rd
harmonics of the drive frequency, and a capacitive impedance matching section, which couples to the drive
coil. The filter schematic is shown in the supplemental data.
2.1.1.6. Receive coil and preamplifier:
The receive coil is a gradiometer where each half is a 27-turn solenoid with an elliptical cross-section (major
diameter =230 mm, minor diameter =160 mm) wound with type-2 Litz wire with 115 strands, 36 AWG
and PFA coating (New England Wire Technologies, Lisbon, NH, USA) . The inductance of the receive coil is
324 µH and resistance is 1.8 Ωat 100 kHz measured with an Agilent 4263B LCR Meter.
The receive coil’s gradiometer design reduces unwanted feedthrough of the 26 kHz drive into the receive
chain. We measure the feedthrough attenuation of the drive and Rx implementation by first placing the
receive coil halfway out of the drive coil to maximize the induced voltage (figure 5, left) and comparing it to
the induced voltage with the Rx coil optimally placed. During this process, the drive coil is powered with
1.67 Apeak and the receive coil voltage is recorded with a floating oscilloscope (RTH1002 Rohde & Schwarz,
Munich, Germany). The coil is then positioned to minimize the induced voltage with drive coil current
increasing to 16.4 Apeak (figure 5, right). The feedthrough attenuation is:
AGrad =VRx,Out
VRx,In ·IDrive,In
IDrive,Out
(1)
where VRx,Out and IDrive,Out is the receive coil voltage and drive coil current with the receive coil halfway
removed. The ‘in’ subscript indicated the coil positioned with the minimum induced voltage. The measured
feedthrough attenuation of the gradiometer is 83 dBc, as seen in figure 5. The receive coil indexes in the drive
7
Phys. Med. Biol. 70 (2025) 015019 E Mattingly et al
Figure 5. Left Column: Receive coil feedthrough with the receive coil halfway out and drive current at 1.67 Apeak. Right Column:
receive coil feedthrough with the receive coil positioned to minimize induced voltage and drive current at 16.4 Apeak.
coil with dovetail wedges, so it can easily be removed and replaced with other application-specific coils for
increased sensitivity (Graeser et al 2020, Chacon-Caldera et al 2024). The process of inserting a receive coil
and fine-tuning its position for minimum feedthrough typically takes less than one minute.
The receive coil is connected to a fully balanced notch filter (Schumacher et al 2020), transformer (1:2
turns ratio), and preamplifier as illustrated in figure 6. Having the shunt section inductors in the receive filter
(i.e. L2/4) and the magnetizing inductance of the transformer be large is important to prevent the reactance
of the receive coil from forming a significant voltage divider with the reactance of these inductors. The
secondary of the transformer has a center-tap to ground to give the bias currents a path to flow and not
saturate the input. The preamplifier is an AD8429 (Analog Devices, Wilmington, Massachusetts)
instrumentation amplifier with a theoretical input noise of 1.2 nV/√Hz and a gain of 200. Following the
first-stage preamplifier, there is a low-pass filter for anti-aliasing (cutoff ≈250kHz), and a single-ended to
differential conversion.
2.1.1.7. Data acquisition system, and software:
The imager control software generates the drive and shift waveforms, controls the motor and gantry, and
ensures accurate synchronization to the digitization of the preamplified SPION signal. It also records data
from other peripherals and monitors such as the gantry rotary position encoder and drive current monitor.
The console operates on a single NI PXIe-1073 that has PXIe-6363 and PXIe-6361 X-Series DAQs to digitize
the received signal and generate the analog drive and shift waveforms, and the same PXIe modules record
digital data from the rotary shaft encoder. It runs on custom-written LabVIEW software (LabVIEW 2021,
Emerson Electric Co., St. Louis, MO). The start of the ADC and DAC is triggered off of each other to ensure
consistent timing.
8
Phys. Med. Biol. 70 (2025) 015019 E Mattingly et al
Figure 6. Receive filter and preamp circuit model. The transformer coupling scales the effective impedance of the Rx coil and
simultaneously mitigates common-mode noises, and gives a path for bias currents to flow. C1/3A/B=170 nF, L1/3A/B=190µH,
C2A/B=9.72nF, L2A/B=1.88 mH, C4A/B=8.66nF, L4A/B=2.12 mH, RG=31.1Ω. The transformer has a turns ratio of 1:2, for
an inductance ratio of 1:4. Following the AD8429, there is low-pass filtering (cutoff ≈250kHz), and single-ended to differential
conversions, which are not shown here for simplicity.
Each image is acquired in a 180-degree rotation of the gantry in 5 s. In that time, 27 projections are
acquired, so the shift frequency is 2.7 Hz (2 projections per period of the triangle wave). Within each of the
projections the data is discretized into 132 equal segments, called ‘read-outs’ (ROs). Each RO contains data
from exactly 35 periods of the drive frequency (1.33 ms) and digitized at 1 mS s−1. The RO time domain data
are Fourier transformed and the complex signal intensity at the 2nd through 9th harmonics is saved and the
other frequency components are discarded. The mean shift currents, which are monitored via the integrated
current monitor on each amplifier, and the gantry position are also recorded during the 1.33 ms RO. The
receive signal components are binned and mapped according the shift amplitude and gantry angle to form
the sinogram consisting of the recorded harmonic amplitude as a function of gantry angle and shift field
amplitude. The operator is supplied with an online view of each image using a simple inverse Radon
transform of the 3rd harmonic data, but the sinograms of the 2nd through 9th harmonics are saved for more
advanced offline reconstructions.
2.2. Image reconstruction
Two different offline 2D reconstruction algorithms are used: an inverse Radon reconstruction, and a
forward-model iterative reconstruction. Both used only the 3rd harmonic (although the higher harmonics
are saved). The 3rd harmonic is valid to use for a 2D acquisition of in-plane SPION distributions due to the
symmetry of the 3rd harmonic sensitivity profile (Mason et al 2022). Each complex projection (column in
the 3rd harmonic sinogram) has a linear complex baseline subtracted off of it by taking a linear fit between
the mean of the first 2 points and mean of the last 2 points and then subtracting this from the measured data.
Essentially, we assume there is no signal at the perimeter of the FOV. The angle for each projection is the
mean measured angle of all the ROs in that projection. As there are occasional spurious spikes in the received
signal (arising from the shift system), these points are identified in the sinogram domain as points more than
three local scaled median absolute deviations away from the local median within a three-element moving
window using MATLAB’s built-in function isoutlier([data],‘movmedian’,3). While we do not know the root
cause of the noise introduced by the shift system, it could result from either insufficient filtering of the
current leads or from intermittent connections in the grounds causing small arcs that have not been found.
The forward model-based (iterative) reconstruction improves on the inverse Radon reconstruction by
including three main pieces of information: The measured gantry angle at each RO, the measured shift
current, and an assumed width of the SPIONs’ point-spread function. The forward model is built with a
simulator in MATLAB 2021 (Mathworks, Natick, MA, USA), and at each time step (corresponding to one
9
Phys. Med. Biol. 70 (2025) 015019 E Mattingly et al
RO) simulator uses the measured shift current, and gradient strength, and measured gantry angle to calculate
the location of the FFL. The ideal simulated FFL is convolved with a 7 mm standard deviation Gaussian
kernel. The 7 mm kernel width was chosen as an approximation of the imager’s native spatial resolution
given the gradient strength and SPION’s saturation magnetization.
The matrix corresponding to this forward model is then inverted with MATLAB’s built-in
preconditioned conjugate gradient solver, pcg(...) and uses 15 iterations, which is picked to balance image
sharpness and artifact amplification. Fewer iterations yield smoother images, and more iterations result in
sharper images with more ringing artifacts.
2.3. Imaging performance tests
2.3.1. Spatial resolution
To evaluate the spatial resolution, we imaged two capillary tubes of 0.5 mgFe mL−1Synomag. The tubes were
2.5 mm diameter, ∼5cm long and were spaced in 1 mm increments between 5 and 9mm apart. A drive
amplitude 5.8 mTpeak was used for the image. For these test images, the shift currents were reduced by
roughly 50% so the FFL moved only ±37mm to provide a 74 mm image FOV. This was done to zoom-in and
increase the sampling density across the phantom. Note that the image resolution is primarily determined by
the SPION’s magnetization curve and gradient strength. The speed of the FFL traversing the FOV (2.7Hz) is
well below the speed at which the relaxation times (on the order of microseconds (Mattingly et al 2024a))
would contribute to the spatial resolution.
The image datasets were reconstructed with both the inverse Radon and forward-model iterative
algorithms. The images from both reconstructions were smoothed with a small 2D Gaussian kernel with a
FWHM of 1.5 mm, which is well below the expected spatial resolution of the system based on the gradient
strength and SPION properties, to suppress noise at higher spatial frequencies than our spatial resolution
would permit.
To measure the spatial resolution, a line was plotted through the middle of the reconstructed images and
the two most prominent peaks were identified in addition to the minimum signal intensity between the two
peaks. The image contrast is defined as:
C=1−Smin
0.5∗(Smax,Left +Smax,Right)(2)
where Cis the dimensionless contrast metric, Smin is the minimum signal between the adjacent peaks, and
Smax,Left/Right are the left and right peaks, respectively. A contrast of greater than 0.5 is considered resolved.
2.3.2. Sensitivity
We measure the detection limit (sensitivity to iron) of the system using an 8-sample dilution series of
Synomag ranging from undiluted (6 mgFe mL−1) to 15.6 µgFe ml−1. The initial concentration was taken from
the manufacturer’s technical datasheet and the following were derived with a serial dilution. For each
concentration, 20 µL is pipetted in a small centrifuge tube to approximate a point-source phantom which is
imaged in 5 s. The FOV for all images is 181 mm and the drive amplitude is 5.8 mTpeak . For the dilution series
imaging, the phantom was carefully placed in a holder at a fixed position in the bore to ensure a repeatable
placement. These data are reconstructed with both reconstruction algorithms and smoothed with a 6 mm
FWHM Gaussian kernel to suppress noise at higher spatial frequencies than could be practically resolved.
This kernel width is chosen based on the results of the spatial resolution tests. The image signal intensity was
recorded from the location of peak signal in the most concentrated image and compared to the noise
standard deviation of all voxels with the FOV of an empty-bore image. After linear fitting to determine the
SNR as a function of iron mass, the iron mass expected to provide an SNR of 5 is defined as the detection
limit.
2.3.3. FOV measurement
The FOV is quantified with a large ‘G’-shaped phantom imaged with a shift current of 250 Apeak to provide
an expected FOV of 19 cm. The phantom was filled with 0.0625 mgFe mL−1Synomag, which is roughly the
concentration of blood assuming 5 mgFe ml−1dose and 65 ml of blood per kg body mass. The circular part
of the ‘G’ is 136 mm in diameter, and that known distance is used to calibrate the total diameter of the FOV.
It was important to have this calibration signal be well contained within the edges of the reconstruction,
because toward the edge of the FOV the peak signal (indicating the center of the lines in the letter) may be
more challenging to identify since the edges of the FOV are used as a baseline in post-processing.
10
Phys. Med. Biol. 70 (2025) 015019 E Mattingly et al
Figure 7. Reconstructions of spatial resolution phantoms. The field of view for each reconstruction is 74 mm diameter. The top
row uses an inverse Radon reconstruction, and the middle row uses the iterative reconstruction. The bottom row contains plots of
the normalized signal magnitude through the center of the image (green solid line =inv. Radon, cyan dashed =iterative recon).
Figure 8. Image contrast (defined as one minus the mean of the peaks divided by the minimum signal between the peaks) for the
reconstructions in figure 7.
2.3.4. Thermal stability
To test the drive coils cooling ability over the course of a typical experiment, we monitored the drive current
stability (and thus field stability) during 35 min of consecutive images with a drive field of 4.6 mTpeak
(∼40 Apeak). Since dissipated power scales with the square of the drive current, these results can be scaled to
different drive currents (and thus drive fields). Upon completion of these images, we removed the drive coil
from the bore and took off the shield and used a thermal camera to measure the temperature of the drive
coils, capacitors, and surrounding structures.
3. Results
3.1. Spatial resolution
Figures 7and 8show the reconstructions of the spatial resolution phantoms and the resulting signal contrast
using the iterative reconstruction as well as the simple inverse Radon reconstruction. As expected, the
iterative reconstruction considerably increases the spatial resolution to 5 mm, whereas the inverse Radon
reconstruction cannot fully resolve the lines until about 7 mm distance between the inner surface lines. The
iterative reconstruction introduces noticeable ringing artifacts while solving the inherently ill-posed
deconvolution problem. The differences in the lengths of the lines are due to discrepancies in filling of the
phantoms.
11
Phys. Med. Biol. 70 (2025) 015019 E Mattingly et al
Figure 9. Left: Images of the dilution series (5 s each) ranging from 120µgFe to 313 ngFe and an empty bore image for comparison
each scaled to the image maximum. Right: The linear regression for the signal versus SPION mass. The 1σand 5σnoise floors are
plotted as horizontal lines.
Figure 10. Left: Photograph of the ‘G’ phantom filled with 0.0625 mg Fe/ml of Synomag. Middle: rendering of the phantom with
red illustrating where the Synomag is filling. The center diameter of the semi-circular component is 136mm and the line width is
4 mm. Right: Inverse radon reconstruction of the data using 3 images (5 s each) averaged together.
3.2. Imager sensitivity
Figure 9shows the imaging results from the Synomag dilution series using the inverse Radon reconstruction.
The resulting images are linear, as expected, with an R2=0.999, and only significantly diverge near the noise
floor due to non-zero baseline signal. The best-fit line intercepts the empty-bore standard deviation (noise
floor with 1σ) at a mass of 150 ngFe, and the 5σline at an Fe mass of 751 ngFe . The iterative reconstruction
(data not shown) has similarly linear results with an R2=0.999, but lower sensitivity as the deconvolution
introduces some signal instability. The iterative reconstruction best-fit line intercepts the 1σhorizontal line
at a mass of 215 ngFe, and the 5σline at a Fe mass of 1077 ngFe.
3.3. FOV
Figure 10 illustrates the achievable FOV of the imager with a ‘G’ phantom extending 136 mm in diameter. By
using this 136 mm known dimension, the entire imager FOV (distance between outermost pixels) is
extrapolated to be 181 mm.
3.4. Thermal stability
Figure 11 shows the thermal images and the corresponding visible-light images following ∼35 min of
continuous imaging. Neither the drive coil nor the capacitors exceeded approximately 40◦C. Notably, the
outer components of the 3D printed structure (white plastic components) are hotter than the drive coil itself
due to their proximity to the copper shield, which is being heated via eddy currents. The amplitude of the
current in the drive coil deviated less than 2% during the 35 min continuous imaging test.
12
Phys. Med. Biol. 70 (2025) 015019 E Mattingly et al
Figure 11. Temperatures as measured following a ∼35 min experiment at 4.5 mTpeak. Top: Capacitors in series with the drive coil.
Bottom: Drive coil wires and former.
4. Discussion
We present the first human-head-sized mechanically rotating FFL-based MPI system. It achieves a spatial
resolution, a temporal resolution (2D image), and a sensitivity that is appropriate for functional human
neuroimaging. It employs the highest localizing gradient to date (1.13 Tm−1in x’) for an MPI system with a
FOV suitable for adult human heads (Lee et al 2006, Rahmer et al 2018, Graeser et al 2019). In addition to
potentially enabling future clinical studies on this prototype instrument, it demonstrates the ability to scale a
number of different technical MPI approaches to human sized imagers, including the use of FFL assemblies
based on continuously rotating permanent magnets and shift coils and high sensitivity gradiometer based
receive coils. By employing other recent developments in human-scale MPI, such as optimized drive coil
wire, mitigating connection-induced distortions, using optimized FFL scanning trajectories, among others
(Top et al 2019, Kretov et al 2024, Mohn et al 2024, Ozaslan et al 2024, Trisnanto et al 2024, Mattingly et al
2024c), the system’s performance could be significantly improved.
4.1. Spatial resolution
We demonstrated spatial resolution of approximately 5mm with an iterative reconstruction and 7 mm with
an inverse Radon reconstruction that does not try to deconvolve the SPION’s point-spread function. Image
resolution is not determined by hardware parameters (such as FFL gradient strength) alone, but is also a
strong function of the SPION used (Bauer et al 2016, Wang et al 2020, Tay et al 2021, Fung et al 2023,
Abdibastami et al 2024), drive field amplitude and frequency (Croft et al 2016, Tay et al 2017,2019b), and
reconstruction deconvolution effort. The iterative reconstruction we demonstrate here could be significantly
improved by utilizing many of the recent developments in MPI reconstruction algorithms (Knopp et al 2019,
Chacon-Caldera et al 2020,2021, Gungor et al 2023, Sanders et al 2024). The scanner’s demonstrated 5 mm
spatial resolution and 5 s temporal resolution was designed for detecting CBV increases associated with
functional brain activation and is comparable to that used in many MRI based functional neuroimaging
applications. While MRI can flexibly achieve higher spatial and temporal resolutions, the majority of
published studies are acquired with a temporal resolution of a few seconds and apply a spatial smoothing
13
Phys. Med. Biol. 70 (2025) 015019 E Mattingly et al
filter ranging from 4–12 mm to the acquired image (Sacchet and Knutson 2013, Triana et al 2020). Our MPI
scanner’s demonstrated spatial resolution would also likely be sufficient for observing the CBV alterations
associated with human traumatic brain injuries.
Applications requiring increased spatial resolution will likely require improvement in the SPION’s
magnetization response or utilization of superconducting FFL and shift coil magnets. Recent developments
in SPION chemistry suggest up to 10 times higher spatial resolution would be possible (Tay et al 2021) and
superconducting magnets are being developed for MPI at large scales with up to 2.5 Tm−1gradient (Le et al
2024) which would increase the spatial resolution another factor of 2.5 from what we have demonstrated.
Future work includes further characterization of the modulation transfer function with varying acquisition
parameters and testing with standard resolution phantoms (e.g. a Derenzo phantom (Cox et al 2016)) to
facilitate comparison to other modalities.
4.2. Imager sensitivity
The Synomag dilution series images demonstrate a SNR =5 (best fit intercept) with 751 ngFe with an
inverse Radon reconstruction and 1077 ngFe with the iterative reconstruction. Thus the practical detection
limit, approximating with one significant figure, is about 1 µgFe for either reconstruction (5 section image,
5.8 mTpeak, 6 mm FWHM smoothing kernel) because of the non-Gaussian nature of the background signal.
This limit of detection (LOD) should be thought of as a point of reference rather than a fixed parameter of
the imager. It is a strong function of scan time (LOD ∝√1/time), drive field amplitude (LOD improves
with higher drive fields), amount of smoothing (LOD improves with more smoothing), SPION used, etc.
Based on the drug labeling for Ferumoxytol (Adkinson et al 2018)), human SPION doses can be expected
to contain 5 mg Fe per kg body mass. Assuming 65 ml blood per kg body mass, the blood concentration of Fe
will be 78 µgFe ml−1. The brain’s gray matter is about 5% blood, so the mass of Fe per tissue volume in gray
matter is expected to be 3.8 µgFe ml−1. The imager produces 6 mm isotropic voxels, or 216 µL, yielding an Fe
mass per voxel of 830 ngFe in cerebral gray matter. For these values, the inverse Radon reconstruction images
should show gray matter with an SNR of about 5 and large blood vessels with an SNR of about 100 in the 5 s
image. A functional task, such as flashing lights, is expected to increase CBV by 25% (Mandeville et al 1998)
and hypercapnia challenges in rodents have induced similar MPI detected blood volume changes (Mason
et al 2023). Based on our sensitivity results, we expect to see this 25% CBV change with a contrast-to-noise
ratio (CNR) above 1 in the 5 section image.
Currently, the scanner’s sensitivity is limited by an unstable background signal originating in the shift coil
amplifier and then coupled into the receive system. Noise measurements with and without the shift system
energized show that this noise source is approximately 10x higher than the receive preamplifier’s noise floor,
motivating future work to mitigate this source with improved filtering of the shift coil currents as they pass
into the shielded room. Similar noise from the drive coil amplifier raises the noise floor by another factor of
two above the preamplifier noise floor. This motivates improving connections and filtering in the drive chain
as well and reducing the pass-band width of the transmit filter (Biwa et al 2006, Krüger et al 2010, Wilkerson
2010).
With these fixes providing a roughly 20x sensitivity boost, we would expect to detect functional CBV
changes during human brain activation with a CNR of roughly 20, exceeding the state of the art achieved by
fMRI. This still does not represent the body-noise dominated case whereby the receive chain noise level is set
by dissipative losses in the body (rather than the preamp or receive coils themselves). To achieve body noise
domination, the noise from the preamplifier and receive coil losses must be addressed. The preamplifier’s
noise floor could be reduced with a number of designs suggested for MPI-tailored preamps. For example, a
voltage noise at or below 150 pV/√Hz has been achieved by using low-noise JFETs such as the 2SK2394(On
Semi, Scottsdale, Arizona) or BF862(NXP, Eindhoven, Netherlands) in a parallelized common-source
configuration (Schmale et al 2010, Graeser et al 2017, Mattingly et al 2022a). Similarly low noise values have
also been achieved with transformer matching or parallel opamp inputs (Zheng et al 2017, Malhotra et al
2020). Due to the large inductance of the receive coils, cascoding the input JFETs (Sedra and Smith 2004,
Huynh 2018) or using other hybrid amplifier designs (Horowitz and Hill 2015) may be necessary to mitigate
the Miller effect capacitance (Sedra and Smith 2004), but this comes at the expense of complexity. These
suggest roughly another order of magnitude lower noise floor could be available. For MRI-based functional
neuroimaging, the noise in the image timeseries is dominated by the ‘physiological noise,’ which originates
from numerous biological functions uncorrelated to the given task under investigation (Triantafyllou et al
2006). These biological processes impose nuisance fluctuations on the large signal baseline, only a small
fraction of which derives from the blood pool. Because MPI detects only the cerebral blood pool and
therefore eliminates much of the signal susceptible to modulation, it is likely less susceptible to these sources
14
Phys. Med. Biol. 70 (2025) 015019 E Mattingly et al
Figure 12. Drive coil current over the course of a ∼35min continuous imaging experiment.
of instability (e.g. respiration artifacts from the magnetization of the air in the lungs, spin history effects,
etc). Thus, the biological signal fluctuations would be primarily CBV fluctuations, which are the source of
valuable functional connectivity information.
4.3. FOV
The FOV shown in figure 10 is sufficient to cover the cross-section of most human brains. This is partially
due to the shift amplifiers which require repair to operate at their specified power, and limits the FOV to
about 83% of the system design. With this repair, the expected FOV is 220 mm, and should comfortably
encompass the head. The imaging reconstructions are also 2D in their current formulation, but this is not an
inherent limitation of the device. To encode a third dimension (the zaxis), there are a few possible
approaches: the subject could be moved in an automated manner, as done in spiral CT, or as in pre-clinical
mechanically rotating FFL devices (Mattingly et al 2022a), a DC-offset current could be added to the drive
coil, or another low-frequency coil could be added to shift the FFL in the zaxis. Any approach to encoding in
the third dimension would have to consider the effects on temporal resolution for each frame of an fMPI
timeseries. This would additionally require improvements to the reconstruction to include 3D information
in the model, which has been done with the SLICE reconstruction on the similar architecture rodent-scale
MPI system (Mason et al 2022), or other 3D reconstruction algorithms (e.g. system matrix (Graeser et al
2019) or X-Space (Goodwill and Conolly 2011)).
4.4. Temporal stability
For time-series imaging, having thermal stability is critical. Figures 11 and 12 show the temperature and
drive coil currents after over 30min of continuous imaging. Over the course of the experiment, the coil
current drifted less than 2%. The drift was generally smooth, slow, and relatively low amplitude compared to
the ∼25%changes expected from brain activation. Because of these features, it is straightforward to separate
it out from the biological signal changes associated with brain activation using a linear model (Mason et al
2023). Altogether, it is unlikely thermal drift will be a prohibitive component in functional imaging
experiments.
While the heating experiments were carried out with 40 Apeak, the temperatures we observed would scale
roughly with the current squared. So, for 50 Apeak , the temperature rises would be about (50/40)2×or 50%
higher, which would still be well below the threshold for damaging components. If needed, active cooling of
the capacitors and the shielding tube could potentially be applied. Additionally, water coolant pressure could
be increased if needed.
Another source of temporal drift is the washout of the SPIONs from circulation. In prior experiments
using Synomag-D the half-life was measured as 48 min (Mason et al 2023), which is consistent with others’
findings (Szwargulski et al 2020, Liu et al 2021). Longer circulations are also possible by tuning the size and
surface properties of the SPIONs. To account for this drift in the analysis, an additional exponential decay
term can be included because the trend is slow and predictable.
15
Phys. Med. Biol. 70 (2025) 015019 E Mattingly et al
4.5. Biological effects of the applied dB/dt
Human MPI will impose a number of different time-varying magnetic fields on the body that may induce
peripheral nerve stimulation (PNS), tissue heating from the specific absorption rate (SAR), and
magneto-stimulation of the retinas. The shift fields induce a magnetic field approximately 100mTpeak over
the head at 2.7 Hz. Due to the low frequency of this applied field, the dB/dtand thus the induced electric
fields in the body are relatively small. The FFL being rotated also causes a dB/dtbut at even lower
frequencies. While there is limited data on the topic, our prior analysis shows that for frequencies below
10 Hz stimulation is unlikely (Mason 2020) although retinal stimulation becomes a concern for frequencies
of 20 Hz (Lovsund et al 1980).
For our drive field operating at 26.3 kHz at up to 7 mTpeak, Saritas et al have shown SAR is unlikely to be a
limiting factor (Saritas et al 2013). The PNS limits in the head are still unclear, but preliminary studies have
shown that the 7 mTpeak may be at or near the limit (Ozaslan et al 2022, Barksdale et al 2024), and future
work is needed to elucidate the safe limits.
5. Conclusions
We demonstrate, for the first time, that a human-scale MPI system with a sufficiently strong gradient
(1Tm−1) to produce clinically-relevant images can be realized with high sensitivity (1 µgFe) for continuous
5 s CBV imaging. We expect this device to be sufficient for seeing the hemodynamic modulations following
brain activity in vivo although the instrument is still far from the expected limits of MPI detection sensitivity
and further mitigation of noise sources will be needed to achieve the full potential of functional MPI. But,
with the appropriate investment in engineering, human brain MPI can potentially provide an order of
magnitude improvement in sensitivity at a useful spatial resolution bringing a new and high-sensitivity
modality to human neuroimaging.
Data availability statement
The data that support the findings of this study are openly available at the following URL/DOI: https://
github.com/EliMattingly22/Data_DesignConstrVal_HumanMPIManuscript.
Acknowledgment
Research reported in this publication was supported by National Institute of Biomedical Imaging and
Bioengineering (NIBIB) of the National Institutes of Health (NIH) under award numbers U01EB025121,
5T32EB1680, and F30MH129062, and by the National Science Foundation (NSF) Graduate Research
Fellowships Program number 1122374. The content is solely the responsibility of the authors and does not
necessarily represent the official views of the National Institutes of Health or the National Science
Foundation. The authors would also like to thank John Drago, Jason Stockmann, Clarissa
Zimmerman-Cooley, Stephen Cauley, who all were available for helpful conversations and provided technical
input along the way.
Conflict of interest
E M and E E M are currently employees of Magnetic Insight, Inc. L L W receives research funding from
Siemens Healthineers. L L W is a co-founder, equity holder and consultant for Neuro42 Inc, a company
pursuing point of care brain MRI.
ORCID iDs
Eli Mattingly https://orcid.org/0000-0002-3177-7619
Erica Mason https://orcid.org/0000-0001-7002-3599
Alex Barksdale https://orcid.org/0000-0001-6533-1079
Frauke H Niebel https://orcid.org/0000-0003-3394-0122
Matthias Graeser https://orcid.org/0000-0003-1472-5988
References
Abdibastami A et al 2024 How size and composition of cobalt doped iron oxide nanoparticle tracers enhance magnetic particle imaging
performance Chem. Mater. 86 84
Aderhold E, Graeser M and Wieprecht H 2023 Modular MPI component testing facility Int. J. Magn. Part. Imaging 954
16
Phys. Med. Biol. 70 (2025) 015019 E Mattingly et al
Adkinson N F, Strauss W E, Macdougall I C, Bernard K E, Auerbach M, Kaper R F, Chertow G M and Krop J S 2018 Comparative safety
of intravenous ferumoxytol versus ferric carboxymaltose in iron deficiency anemia: a randomized trial Am. J. Hematol. 93 683–90
Barksdale A, Ferris N, Mattingly E, Sliwiak M, Guerin B, Wald L, Davids M and Klein V 2024 Measured PNS thresholds in a human
head MPI solenoid from 200 Hz to 88.1 kHz Int. J. Magn. Part. Imaging 10
Bauer L M, Situ S F, Griswold M A and Samia A C S 2016 High-performance iron oxide nanoparticles for magnetic particle
imaging-guided hyperthermia (hMPI) Nanoscale 812162–9
Biwa S, Hiraiwa S and Matsumoto E 2006 Experimental and theoretical study of harmonic generation at contacting interface Ultrasonics
44 e1319–22
Chacon-Caldera J, Lehr H, Sajjamark K and Franke J 2020 Eigen-reconstructions: a closer look into the system matrix Int. J. Magn. Part.
Imaging 672
Chacon-Caldera J, Lehr H, Sajjamark K and Franke J 2021 Enhancing spatial resolution in magnetic particle imaging using
eigen-reconstructions: opportunities and limitations Int. J. Magn. Part. Imaging 73
Chacon-Caldera J, Mattingly E, Sliwiak M, Barksdale A C, Niebel F H and Wald L L 2024 A receive insert for non-human primate
functional MPI (fMPI) Int. J. Magn. Part. Imaging 10 11
Cox B L, Graves S A, Farhoud M, Barnhart T E, Jeffery J J, Eliceiri K W and Nickles R J 2016 Development of a novel linearly-filled
Derenzo microPET phantom Am. J. Nucl. Med. Mol. Imaging 6199–204
Croft L R, Goodwill P W, Konkle J J, Arami H, Price D A, Li A X, Saritas E U and Conolly S M 2016 Low drive field amplitude for
improved image resolution in magnetic particle imaging Med. Phys. 43 424–35
DeYoe E A, Bandettini P, Neitz J, Miller D and Winans P 1994 Functional magnetic resonance imaging (FMRI) of the human brain J.
Neurosci. Methods 54 171–87
Erbe M, Knopp T, Sattel T F, Biederer S and Buzug T M 2011 Experimental generation of an arbitrarily rotated field-free line for the use
in magnetic particle imaging Med. Phys. 38 5200–7
Ergor M and Bingolbali A 2022 Field-free line magnetic particle imaging magnet design using nested halbach cylinders IEEE Magn. Lett.
13 1–4
Ferkovic L and Ilic D 2007 Dependence of mutual inductance of a precise Rogowski coil on the primary conductor position 15th IMEKO
TC4 Symp. on Novelties in Electrical Measurements and Instrumentation vol 58 pp 122–8
Fung K L B et al 2023 First superferromagnetic remanence characterization and scan optimization for super-resolution magnetic
particle imaging Nano Lett. 23 1717–25
Gleich B and Weizenecker J 2005 Tomographic imaging using the nonlinear response of magnetic particles Nature 435 1214–7
Goodwill P W and Conolly S M 2010 The X-space formulation of the magnetic particle imaging process: 1-D signal, resolution,
bandwidth, SNR, SAR and magnetostimulation IEEE Trans. Med. Imaging 29 1851–9
Goodwill P W and Conolly S M 2011 Multidimensional X-space magnetic particle imaging IEEE Trans. Med. Imaging 30 1581–90
Graeser M et al 2017 Towards picogram detection of superparamagnetic iron-oxide particles using a gradiometric receive coil Sci. Rep.
76872
Graeser M et al 2019 Human-sized magnetic particle imaging for brain applications Nat. Commun. 10 1936
Graeser M, Liebing T, Szwargulski P, Förger F, Thieben F, Ludewig P and Knopp T 2020 Organ specific mouse head coil for improved
image quality in magnetic particle imaging Int. J. Magn. Part. Imaging 61–3
Greiner C, Rückert M A, Kampf T, Behr V C and Vogel P 2022 Traveling wave MPI utilizing a field-free line Int. J. Magn. Part. Imaging
88–10
Gungor A, Askin B, Soydan D A, Saritas E U, Top C B and Cukur T 2023 A denoiser scaling technique for plug-and-play MPI
reconstruction Int. J. Magn. Part. Imaging 941
Horowitz P and Hill W 2015 The Art of Electronics 3rd edn (Cambridge University Press)
Huettel S A, Song A W and McCarthy a G 2014 Functional Magnetic Resonance Imaging 3rd edn (Oxford University Press)
Huynh Q 2018 Ultra low noise preamplifier design for magnetic particle imaging Master’s Thesis EECS Department, University of
California, Berkeley, Berkeley, California, USA
Khandhar P, Keselman A, Kemp P J, Ferguson S M, Goodwill R W, Conolly P M and Krishnan M 2017 Evaluation of PEG-coated iron
oxide nanoparticles as blood pool tracers for preclinical magnetic particle imaging Nanoscale 91299–306
Kluth T 2018 Mathematical models for magnetic particle imaging Inverse Problems 34 083001
Knopp T, Gdaniec N, Rehr R, Graeser M and Gerkmann T 2019 Correction of linear system drifts in magnetic particle imaging Phys.
Med. Biol. 64 125013
Knopp T, Sattel T F, Biederer S and Buzug T M 2010a Field-free line formation in a magnetic field J. Phys. A: Math. Theor. 43 012002
Knopp T, Sattel T F, Biederer S, Rahmer J, Weizenecker J, Gleich B, Borgert J and Buzug T M 2010b Model-based reconstruction for
magnetic particle imaging IEEE Trans. Med. Imaging 29 12–18
Kretov E, Scheel J-P, Mirzojan L, Sevecke F and Graeser M 2024 Power-optimized drive field coils for human brain magnetic particle
imaging Int. J. Magn. Part. Imaging 10 02
Krüger M, Nils F N and Reichl H 2010 Intermodulation distortion as indicator for interconnect degradation 2010 Proc. 60th Electronic
Components and Technology Conf. (ECTC) pp 477–83
Le T-A, Bui M P, Hadadian Y, Gadelmowla K M, Oh S, Im C, Hahn S and Yoon J 2024 Towards human-scale magnetic particle imaging:
development of the first system with superconductor-based selection coils IEEE Trans. on Medical Imaging pp 1–1
Lee J-h, Shin S-j H and Istook C L 2006 Analysis of human head shapes in the United States Int. J. Human Ecol. 777–83
Liu S, Chiu-Lam A, Rivera-Rodriguez A, DeGroff R, Savliwala S, Sarna N and Rinaldi-Ramos C M 2021 Long circulating tracer tailored
for magnetic particle imaging Nanotheranostics 5348–61
Lovsund P, Oberg P and Nilsson S E G 1980 Magneto- and electrophosphenes: a comparative study Med. Biol. Eng. Comput. 18 758–64
Ludewig P et al 2017 Magnetic particle imaging for real-time perfusion imaging in acute stroke ACS Nano 11 10480–8
Malhotra A, Schwegmann H, Schumacher J, Chen X and Buzug T M 2020 Fully differential low noise amplifier for MPI/MPS Int. J.
Magn. Part. Imaging 61–3
Mandeville J B, Marota J J, Kosofsky B E, Keltner J R, Weissleder R, Rosen B R and Weisskoff R M 1998 Dynamic functional imaging of
relative cerebral blood volume during rat forepaw stimulation Magn. Reson. Med. 39 615–24
Mason E E 2020 Magnetic particle imaging technology for clinical intraoperative breast cancer margin assessment and functional brain
imaging PhD Thesis Massachusetts Institute of Technology
Mason E E et al 2023 Functional magnetic particle imaging (fMPI) of cerebrovascular changes in the rat brain during hypercapnia Phys.
Med. Biol. 68 175032
Mason E E et al 2024 Preliminary results: large bore clinical MPI system imaging human head-sized FOVs Int. J. Magn. Part. Imaging 10
17
Phys. Med. Biol. 70 (2025) 015019 E Mattingly et al
Mason E E, Cauley S F, Mattingly E, Sliwiak M and Wald L L 2022 Side lobe informed center extraction (SLICE): a projection-space
forward model reconstruction for a 2D imaging system Int. J. Magn. Part. Imaging 823
Mason E E, Cooley C Z, Cauley S F, Griswold M A, Conolly S M and Wald L L 2017 Design analysis of an MPI human functional brain
scanner Int. J. Magn. Part. Imaging 31–13
Mattingly E 2024 Design, construction, and validation of magnetic particle imaging systems for rodent, primate, and human functional
neuroimaging. Thesis Massachusetts Institute of Technology
Mattingly E, Barksdale A C, Sliwiak M, Chacon-Caldera J, Mason E E and Wald L L 2024a Open-source device for high sensitivity
magnetic particle spectroscopy, relaxometry and hysteresis loop tracing Rev. Sci. Instrum. 95 063706
Mattingly E, Mason E E, Herb K, Sliwiak M, Drago J, Graeser M and Wald L L 2022a A sensitive, stable, continuously rotating FFL MPI
system for functional imaging of the rat brain Int. J. Magn. Part. Imaging 801
Mattingly E, Sliwiak M, Chacon-Caldera J, Barksdale A, Niebel F, Mason E and Wald L 2024b A human-scale magnetic particle imaging
system for functional neuroimaging Int. J. Magn. Part. Imaging 10 16
Mattingly E, Sliwiak M, Chacon-Caldera J, Barksdale A and Wald L 2024 Optimization of wall thickness for water-cooled hollow
conductor drive coils in human-sized MPI Int. J. Magn. Part. Imaging 10 24
Mattingly E, Sliwiak M, Drago J M, Mason E E, Graeser M and Wald L L 2022b A drive filter design for MPI with harmonic notching
and selective damping Int. J. Magn. Part. Imaging 873
Mohn F, Forger F, Thieben F, Moddel M, Schmale I, Knopp T and Graeser M 2024 Resonant inductive coupling network for
human-sized magnetic particle imaging Rev. Sci. Instrum. 95 044701
Nomura K, Washino M, Matsuda T, Seino S, Nakagawa T, Kiwa T and Kanemaru M 2024 Development of human head size magnetic
particle imaging system Int. J. Magn. Part. Imaging 10 1
Orendorff R et al 2017 First in vivo traumatic brain injury imaging via magnetic particle imaging Phys. Med. Biol. 62 3501–9
Ozaslan A A, Babaloo R, Atalar E and Saritas E U 2024 Minimizing induced electric fields in human head-size MPI Systems Int. J. Magn.
Part. Imaging 10 28
Ozaslan A A, Utkur M, Canpolat U, Tuncer M A, Oguz K K and Saritas E U 2022 PNS limits for human head-size MPI systems:
preliminary results Int. J. Magn. Part. Imaging 82203028
Pagan J, McDonough C, Vo T and Tonyushkin A 2021 Single-sided magnetic particle imaging device with field-free-line geometry for
in vivo imaging applications IEEE Trans. Magn. 57 1–5
Rahmer J, Stehning C and Gleich B 2018 Remote magnetic actuation using a clinical scale system PLoS One 13 e0193546
Rahmer J, Weizenecker J, Gleich B and Borgert J 2009 Signal encoding in magnetic particle imaging: properties of the system function
BMC Med. Imaging 94
Ramboz J 1996 Machinable Rogowski coil, design and calibration IEEE Trans. Instrum. Meas. 45 511–5
Sacchet M D and Knutson B 2013 Spatial smoothing systematically biases the localization of reward-related brain activity NeuroImage
66 270–7
Sanders T, Mason E, Konkle J and Goodwill P 2024 Multi-harmonic gridded 3D deconvolution (MH3D) for image reconstruction in
MPI: MH3D Int. J. Magn. Part. Imaging 10
Saritas E U, Goodwill P W, Zhang G Z and Conolly S M 2013 Magnetostimulation limits in magnetic particle imaging IEEE Trans. Med.
Imaging 32 1600–10
Sattel T F, Knopp T, Biederer S, Gleich B, Weizenecker J, Borgert J and Buzug T M 2009 Single-sided device for magnetic particle
imaging J. Phys. D: Appl. Phys. 42 022001
Schmale I, Gleich B, Borgert J and Weizenecker J 2010 Jfet noise modelling for MPI receivers Magn. Nanopart. 148–53
Schumacher J, Malhotra A, Grafe K and Buzug T M 2020 Highly symmetric filter for a fully differential receive chain Int. J. Magn. Part.
Imaging 61–3
Sedra A S and Smith K C 2004 Microelectronic Circuits (The Oxford Series in Electrical and Computer Engineering ) 5th ed. (Oxford
University Press)
Shliomis M 1972 Effective viscosity of magnetic fluids J. Exp. Theor. Phys. 34 1291–4
Szwargulski S et al 2020 Monitoring intracranial cerebral hemorrhage using multicontrast real-time magnetic particle imaging ACS
Nano 14 13913–23
Tay Z W et al 2021 Superferromagnetic nanoparticles enable order-of-magnitude resolution & sensitivity gain in magnetic particle
imaging Small Methods 52100796
Tay Z W, Hensley D W, Chandrasekharan P, Zheng B and Conolly S M 2019b Optimization of drive parameters for resolution,
sensitivity and safety in magnetic particle imaging IEEE Trans. Med. Imaging 39 1–1
Tay Z W, Hensley D W, Vreeland E C, Zheng B and Conolly S M 2017 The relaxation wall: experimental limits to improving MPI spatial
resolution by increasing nanoparticle core size Biomed. Phys. Eng. Express 3035003
Tay Z W, Hensley D, Ma J, Chandrasekharan P, Zheng B, Goodwill P and Conolly S 2019a Pulsed excitation in magnetic particle imaging
IEEE Trans. Med. Imaging 38 2389–99
Thieben F, Foerger F, Mohn F, Hackelberg N, Boberg M, Scheel J-P, Moddel M, Graeser M and Knopp T 2024 System characterization of
a human-sized 3D real-time magnetic particle imaging scanner for cerebral applications Commun. Eng. 347
Top C B D, Ilbey S and Guven H E 2017 Electronically rotated and translated field-free line generation for open bore magnetic particle
imaging: Med. Phys. 44 6225–38
Top C B, Gungor A, Ilbey S and Guven H E 2019 Trajectory analysis for field free line magnetic particle imaging Med. Phys. 46 1592–607
Triana A M, Glerean E, Saramaki J and Korhonen O 2020 Effects of spatial smoothing on group-level differences in functional brain
networks Netw. Neurosci. 4556–74
Triantafyllou C, Hoge R D and Wald L L 2006 Effect of spatial smoothing on physiological noise in high-resolution fMRI NeuroImage
32 551–7
Trisnanto S B, Kasajima T, Shibuya T and Takemura Y 2024 Gradiometric human head coil-coupled magnetoresistive sensor for
sensitivity improvement in magnetic particle imaging Int. J. Magn. Part. Imaging 10
Vogel P, Ruckert M A, Greiner C, Gunther J, Reichl T, Kampf T, Bley T A, Behr V C and Herz S 2023 iMPI: portable human-sized
magnetic particle imaging scanner for real-time endovascular interventions Sci. Rep. 13 10472
Wang Q, Ma X, Liao H, Liang Z, Li F, Tian J and Ling D 2020 Artificially engineered cubic iron oxide nanoparticle as a
high-performance magnetic particle imaging tracer for stem cell tracking ACS Nano 14 2053–62
Weber M, Beuke J, Gladiss A v, Grafe K, Vogel P, Behr V C and Buzug T M 2018 Novel field geometry using two halbach cylinders for
FFL-MPI Int. J. Magn. Part. Imaging 41811004
Weizenecker J, Gleich B and Borgert J 2008 Magnetic particle imaging using a field free line J. Phys. D: Appl. Phys. 41 105009
18
Phys. Med. Biol. 70 (2025) 015019 E Mattingly et al
Weizenecker J, Gleich B, Rahmer J, Dahnke H and Borgert J 2009 Three-dimensional real-time in vivo magnetic particle imaging Phys.
Med. Biol. 54 L1
Wilkerson J R 2010 Passive intermodulation distortion in radio frequency communication systems PhD Thesis North Carolina State
University, United States – North Carolina
Wu L et al 2019 A review of magnetic particle imaging and perspectives on neuroimaging Am. J. Neuroradiol. 40 206–12
Yoshida T, Kamei Y, Takemura Y, Nagano T, Ide A and Sasayama T 2024 Development of magnetic particle imaging modules using
high-Tc superconducting coils Int. J. Magn. Part. Imaging 10 38
Yu E Y et al 2017 Magnetic particle imaging for highly sensitive, quantitative and safe in vivo gut bleed detection in a murine model ACS
Nano 11 12067–76
Zheng B, Goodwill P W, Dixit N, Xiao D, Zhang W, Gunel B, Lu K, Scott G C and Conolly S M 2017 Optimal broadband noise matching
to inductive sensors: application to magnetic particle imaging IEEE Trans. Biomed. Cir. Syst. 11 1041–52
Zigrang D J and Sylvester N D 1985 A review of explicit friction factor equations J. Energy Resour. Technol. Trans. ASME 107 280–3
19
