Enhanced performance of a single-sided MPI scanner with a hybrid surface-volume excitation coil

Abstract

Objective.

Magnetic Particle Imaging (MPI) is a tracer-based imaging modality featuring high sensitivity and no ionizing radiation, offering strong potential for cancer detection and vascular imaging. However, clinical translation remains challenging due to the difficulty of scaling closed-bore MPI scanners. Single-sided MPI scanners offer a promising alternative by placing all hardware on one side of the subject, enabling open access and compatibility with larger anatomies. However, they suffer from magnetic field inhomogeneity and reduced sensitivity. In this work, we demonstrate a next-generation single-sided MPI scanner with targeted hardware and software upgrades to address these limitations.

Approach.

Enhancements to our single-sided field-free line (FFL) MPI scanner include a new hybrid surface-volume excitation coil that generates a stronger, more homogeneous excitation field, enabling improved sensitivity across larger imaging regions and a third selection coil. Additionally, a new FFL trajectory correction algorithm was implemented to achieve linear spatial encoding. Scanner performance was evaluated through phantom studies assessing sensitivity, spatial resolution, FFL trajectory linearization, and imaging field of view (FOV).

Main results.

SPIO dilution studies demonstrated a detection threshold of 2.0 μg of iron. Spatial resolution was evaluated using a two-rod phantom with 1 mm diameter rods spaced 4 mm apart, which were clearly resolved. The flat FFL control algorithm enabled linearized spatial encoding, confirmed through multi-plane imaging of a triple-rod phantom. Imaging of an anatomical mouse phantom and a spiral phantom is demonstrated across the entire 8 cm FOV, a twofold increase in FOV.

Significance.

The demonstrated imaging capabilities meet or exceed those of other human-focused MPI scanners, representing a substantial advance for single-sided scanner design. The scanner’s performance and accessibility suggest suitability for clinical applications such as breast imaging and near-surface tumor detection.

1. Introduction

Magnetic Particle Imaging (MPI) is an emerging tracer-based imaging modality that enables quantitative, high-contrast visualization of superparamagnetic iron oxide nanoparticles (SPIOs) without the use of ionizing radiation Gleich and Weizenecker (2005). Its high sensitivity, fast temporal resolution, and safety, due to both the absence of ionizing radiation and the low toxicity of SPIO tracers, make it well-suited for clinical applications such as vascular imaging Weizenecker et al. (2009), cancer detection Yu et al. (2017), and cell tracking Zheng et al. (2015, 2016); Bulte et al. (2015). These features position MPI as a promising alternative or complement to established modalities such as PET, CT, and MRI, especially in scenarios requiring high specificity.

Traditional MPI systems use closed-bore geometries, which enable uniform magnetic fields across the imaging volume through the use of Maxwell coil configurations. This uniformity supports consistent spatial resolution and sensitivity, making closed-bore scanners the dominant design in preclinical MPI. Notable implementations include commercial systems from Magnetic Insight and Bruker, as well as research-grade custom systems Goodwill et al. (2012); Buzug et al. (2012); Saritas et al. (2013). While these systems perform well in preclinical settings, scaling to whole-body or large-subject imaging is challenging due to high power requirements Borgert et al. (2013). As a result, clinical development has focused on smaller targets, such as the head or extremities, where reduced bore size lowers the power needed to generate sufficient field strength. Significant research efforts are now directed toward head-only scanners designed for brain imaging applications, including both anatomical Graeser et al. (2019); Thieben et al. (2024); Le et al. (2024); Wei et al. (2024) and functional MPI systems Mattingly et al. (2025). Closed-bore configurations have also been adapted for other anatomies, including a scanner specifically designed for real-time imaging of the leg during vascular procedures Vogel et al. (2023); Hartung et al. (2025). However, these designs are inherently limited in anatomical flexibility and are not well-suited for regions that are part of a whole body.

Single-sided MPI scanners eliminate these constraints by positioning all hardware to a single side of the imaging volume. This unobstructed configuration makes single-sided scanners ideal for applications involving smaller regions of the whole body, such as in breast imaging. However, this comes with the inherent challenge of inhomogeneous magnetic fields resulting in spatially varying sensitivity and resolution, as well as limited imaging depths. Since the original proposal of this scanner geometry Sattel et al. (2009), several groups have developed or improved upon the single-sided topology Kaethner et al. (2015); Gräfe et al. (2016); Wang et al. (2025). Among these, our single-sided field-free line (FFL) MPI scanner Tonyushkin (2017); Pagan et al. (2021) stands out as one of the highest performing systems of its kind to date McDonough et al. (2024). It has demonstrated the highest sensitivity and spatial fidelity compared to other single-sided scanners, enabled by FFL-based spatial encoding, which enhances sensitivity by sampling a larger instantaneous SPIO volume than field-free point systems Weizenecker et al. (2008); Konkle et al. (2013), and a surface receive gradiometer geometry that provides high sensitivity with effective feedthrough cancellation McDonough et al. (2022b). Despite these advantages, the initial version of the scanner faced limitations common to single-sided systems. In particular, reduced sensitivity at the imaging height led to extended scan times, and field inhomogeneity resulted in curved FFL trajectories that distorted spatial encoding, resulting in artifacts of the reconstructed images and limiting the usable field of view (FOV).

In this work, we present significant performance enhancements to a single-sided FFL MPI scanner that address these limitations to further advance the scanner to clinical viability. The improvements include both hardware and software upgrades: a new hybrid surface-volume excitation coil designed to accommodate a 185 mm rotational table, an additional selection coil, and a new FFL trajectory control algorithm that enables electronically flattened scans across extended FOVs. The performance of the scanner is validated through phantom studies that demonstrate imaging an 8 cm FOV with improved homogeneity, a sensitivity of 2.0 μg of iron, and a spatial resolution of 4 mm. Due to the single-sided nature of our design, the system can accommodate arbitrarily large subjects above the surface-volume excitation coil or subjects up to 19.5 cm in diameter within the coil. This makes our scanner comparable to human-scale MPI systems, which typically feature 20 cm bore diameters, rather than to smaller preclinical systems. The achieved sensitivity surpasses many of the human MPI scanners Graeser et al. (2019); Vogel et al. (2023); Wei et al. (2024) and approaches the highest sensitivity reported for human systems Mattingly et al. (2025). Similarly, the demonstrated spatial resolution compares favorably to other human MPI systems, which typically operate at lower gradient strength (0.2-0.36 T/m), in contrast to the 1.0 T/m gradient demonstrated in this work.

2. Material and Methods

In this section, we provide the details on the hardware of the single-sided FFL scanner and the measurement protocols used for each study.

2.1. MPI System Hardware

The single-sided MPI scanner used in this study is a modified version of the system previously described in McDonough et al. (2024). The current system comprises six electromagnetic coils: three embedded racetrack-shaped selection coils housed within a custom water-cooled enclosure, and three surface-mounted coils, including the receive coil, cancellation gradiometer, and a newly added surface-volume excitation coil.

The two uppermost embedded racetrack coils function as primary selection coils, producing a static magnetic field gradient that creates an FFL oriented along the y-axis. A third embedded racetrack coil, originally utilized as an excitation coil McDonough et al. (2024), is now repurposed as an auxiliary selection coil. This modification introduces an additional degree of freedom, enabling full control over the FFL position in the XZ plane. As a result, the system can maintain a flat FFL trajectory at a fixed height of 17 mm above the scanner surface and extend the usable FOV.

Signal detection is performed by a circular receive coil located at the center of the scanner surface. A second, laterally offset circular coil functions as a cancellation gradiometer to suppress background signal from the excitation field McDonough et al. (2022b). Both coils are easily replaceable, enabling flexible adaptation for specific applications.

New to this iteration of the system is the surface-volume excitation coil, which generates a 25 kHz AC magnetic field to drive SPIO oscillations. This racetrack-shaped coil surrounds the 185 mm subject table shown in Fig.1(a), providing a stronger and more homogeneous excitation field across the imaging FOV compared to the original embedded excitation coil. Its geometry also provides improved coupling to the surface-mounted receive and gradiometer coils, further enhancing signal quality McDonough et al. (2022b). To allow water cooling, high thermal conductivity H2 PFA tubes were wrapped around the coil (visible as gray tubing in Fig. 1(a)).

Figure 1.

(a) Photograph of the single-sided FFL MPI scanner. Visible components include the new surface-volume excitation coil surrounded by the water-cooling tubing, the surface-mounted gradiometer, the water-cooled enclosure housing the embedded selection coils, and a mouse phantom positioned on the 185 mm subject table. (b) 3D rendering of the scanner showing the three embedded racetrack-shaped selection coils, the surface-mounted receive and gradiometer coils, and the surface-volume excitation coil.

2.2. Surface-Volume Excitation

A major hardware upgrade in the current scanner design is the addition of the new organ-specific hybrid surface-volume excitation coil, which replaces the embedded excitation coil used in prior work McDonough et al. (2024). The new geometry improves the strength and uniformity of the excitation field in the imaging FOV while maintaining compatibility with the surface receive and gradiometer coils McDonough et al. (2022b).

The excitation coil is designed to accommodate and partially enclose a body region for a specific application, e.g., breast cancer diagnostics. For breast imaging applications, accessibility is not compromised because the anatomy naturally extends within the coil. If imaging such anatomies becomes desirable, separate surface–volume coils can be developed for those targets to maintain accessibility. In the current setup, the coil features a racetrack geometry with an inner bend diameter of 195 mm and a straight segment length of 70 mm along the long axis, providing clearance for the subject table. The overall outer dimensions are 315 mm in length and 245 mm in width, and the coil has a total height of 28.5 mm (see Fig. 2). It was custom fabricated by winding 14 AWG Litz wire around a 3D printed PLA former, and glued together using a clear adhesive. The coil consists of eight layers with ten turns per layer. The layers are connected in a series-parallel configuration: four sets of two layers are connected in series, and those four series elements are then connected in parallel. This configuration reduces the coil’s inductance and supports compatibility with the custom low-pass excitation filter. The measured electrical properties at 25 kHz are 141.1 μH inductance and 417 mΩ resistance, as determined using an LCR meter.

Figure 2.

Schematic of the surface-volume excitation coil. The coil has a racetrack geometry with outer dimensions of 315 mm × 245 mm and a winding width of 25 mm. The inner diameter measures 195 mm with a straight segment length of 70 mm along the long axis, providing clearance for the 185 mm subject table. Markers indicate positions of geometric symmetry defined by the coil layout, where the receive and gradiometer coils are placed. This symmetry results in nearly identical coupling to the excitation field and enables improved gradiometer cancellation. The accompanying contour plot shows the simulated B_z excitation field at 17 mm above the scanner surface. The field varies by 27% across the 8 cm FOV with an average amplitude of 0.109 mT/A, and a central value of 0.102 mT/A.

The coil’s geometry enables symmetric placement of the receive and gradiometer coils relative to the excitation coil, as shown in Fig. 2. This spatial symmetry results in nearly identical coupling to both receive coils, substantially improving gradiometer cancellation from 41 dB with the embedded excitation coil to 55 dB with the surface-volume excitation coil McDonough et al. (2022b). Mounting the excitation coil directly on the scanner surface allows it to surround the imaging region, increasing both field homogeneity and field strength due to its proximity to the imaging plane (17 mm above the scanner surface). In this plane, the coil produces an average magnetic field of 0.109 mT/A across the 8 cm FOV, with a variation of 27% Fig. 2. At the center of the FOV, the field per unit current is 0.102 mT/A. In this work, the coil was driven with up to I_RMS = 29 A, corresponding to an excitation field of 2.96 mT at the imaging height. This compares favorably to the previous embedded excitation coil, which produced only 0.58 mT at the same location.

2.3. Low-Pass Excitation Filter

A custom low-pass excitation filter was developed to deliver high-amplitude, harmonically pure (25 kHz), and thermally stable current to the surface-volume excitation coil. The design is based on a topology proposed in Mattingly et al. (2022), which was optimized specifically for thermal stability. This is an important concern in MPI, as thermal fluctuations in the excitation current can lead to baseline drift and inconsistencies in the background signal. These effects are particularly problematic in systems that rely on background subtraction during reconstruction, a technique that assumes a temporally stable background. Even small instabilities in the excitation waveform can introduce subtraction errors that degrade image quality Mattingly et al. (2022).

The filter circuit, shown in Fig.3, consists of a sequence of LC stages tuned primarily to the 25 kHz excitation frequency. Its role is to pass the excitation frequency and suppress harmonics that would otherwise distort the excitation waveform. The architecture combines series and parallel resonant components: series LC stages provide a low-impedance path for the excitation frequency (25 kHz), while parallel stages shunt higher harmonics to ground. A single-notch filter tuned to 75 kHz is included to suppress the third harmonic explicitly, as our system employs narrowband signal detection centered on the third harmonic. Any background at this frequency would interfere with SPIO signal detection and reduce system sensitivity. Thus, the filter is designed with decreasing impedance along its length. The early stages operate in a voltage-dominant regime, reducing current and thus limiting heating in the early filter elements. Toward the end of the filter, the impedance decreases, enabling higher current delivery to the excitation coil, which in turn produces a stronger excitation field.

Figure 3.

(a) Schematic of the multi-stage resonant filter for our excitation coil. The circuit is divided into five functional stages, highlighted by background color. Red: AE Techron 7548 power amplifier. Blue: Double-tuned bandpass filter with a shunt resonator, centered at 25 kHz. Green: Notch filter tuned to suppress the third harmonic at 75 kHz. Purple: Inductor capacitor pair tuned to 25 kHz to optimize power transfer to the excitation coil. Yellow: Excitation coil with resonance capacitors for impedance matching. (b) Experimentally measured voltage transfer function of the full circuit, acquired using a network analyzer. The black line shows voltage magnitude in dB, and the blue dashed line shows phase in degrees. The red dot marks the resonance peak at 25 kHz.

All inductors were hand-wound using Litz wire to reduce AC resistance and were constructed on air-core formers to avoid the generation of unwanted higher harmonics. High-power capacitors (Celem, Israel) were used in each resonant stage, and were precision-tuned with parallel film capacitors to achieve final capacitance values within 0.1% of the target value. The filter was tuned to match the 141.1 μH impedance of our excitation coil. This inductance was deliberately chosen to ensure the feasibility of constructing the resonant filter elements, particularly the custom inductors, which must have comparable inductance to the load. In previous scanner iterations, the maximum excitation current was limited to 5 A due to heating in the filter. In contrast, the present design supports stable operation at I_RMS = 29 A, yielding an excitation field of 2.96 mT at a height of 17 mm above the scanner surface. The thermal limitation has now shifted to the excitation coil itself, marking a substantial improvement in system performance. In addition to improved current handling, the filter achieves 76.0 dB attenuation at 75 kHz (the third harmonic) and 13.5 dB attenuation at 25 kHz (the fundamental). Aside from the upgraded low-pass filter and new surface–volume excitation coil, the remaining excitation signal chain is identical to that described in McDonough et al. (2024), consisting of a DS360 function generator (Stanford Research Systems); an amplifier Techron 7548, and the low-pass filter connected to the excitation coil. The receive chain is unchanged from the previous system, with the receive coil connected to an SR860 lock-in amplifier (Stanford Research Systems).

2.4. FFL Trajectory Control Algorithm

Precise electronic control of the FFL trajectory is critical for MPI, as it defines the region of signal sensitivity and enables spatial encoding. In closed-bore MPI systems, the FFL is typically formed by symmetric selection fields with well-understood behavior. In contrast, single-sided topologies produce inherently inhomogeneous magnetic fields, making it nontrivial to position the FFL at arbitrary locations or with consistent gradient strength. This lack of linearity complicates both image acquisition and reconstruction. To address this issue, we developed an algorithm that computes the currents required to place the FFL at any desired location, with a specified gradient magnitude. This approach enables full electronic control of both FFL position and spatial resolution across the FOV, and forms the basis for our imaging protocol. The comparison between encodings with and without the correction algorithm is shown in Figure 4.

Figure 4.

Demonstration of FFL trajectory control in the XZ plane. The contour plot shows the magnitude of the static magnetic field in the XZ plane when the FFL is formed at the center of the FOV, 17 mm above the scanner surface. Two trajectories are overlaid: the white dashed line shows the uncorrected FFL trajectory obtained using only two selection coils, which causes the FFL to drop by 7 mm at ±20 mm from the center. This height variation previously limited the usable FOV to 40 mm. The yellow dashed line shows the corrected FFL trajectory generated by the new three-coil algorithm, which maintains a constant height across the entire scan range. This correction enables a linear FFL trajectory and supports imaging over a larger FOV.

Our system employs three independent selection coils; thus, the total magnetic field they generate is given by the superposition of each:

$$B_{tot}(x,z)=I_1{P}_1(x,z)+I_2{P}_2(x,z)+I_3{P}_3(x,z)$$

Here, I_i denotes the current in selection coil i, and P_i(x, z) denotes the magnetic field produced from selection coil i per unit current at the central XZ plane, which we measured experimentally. Because the scanner generates an FFL along the y-axis and the selection coils are much longer than the imaging region, we assume translational invariance in the y-direction. This simplifies the 3D field problem (x,y,z) to a 2D plane in (x,z), allowing a slice-wise approach to field control.

To place the FFL at a specific position r_FFL = (x_FFL, z_FFL), the condition is that the magnetic field vanishes at that position:

$$B_{tot}(r_{FFL})=I_1{P}_1(r_{FFL})+I_2{P}_2(r_{FFL})+I_3{P}_3(r_{FFL})=0$$

Since the magnetic field in the y-direction is negligible, due to our FFL topology, this yields two scalar equations:

$$\begin{gathered} I_1{P}_{1x}(r_{FFL})+I_2{P}_{2x}(r_{FFL})+I_3{P}_{3x}(r_{FFL})=0 \\ {I}_1{P}_{1z}(r_{FFL})+I_2{P}_{2z}(r_{FFL})+I_3{P}_{3z}(r_{FFL})=0 \end{gathered}$$

This is an underdetermined system with two equations and three unknowns. To resolve this, we define the relative current ratios $\alpha =I_2/I_1$ and $\beta =I_3/I_1$, assuming $I_1\ne 0$. Dividing both equations by I₁ yields:

$$\begin{gathered} P_{1x}(r_{FFL})+\alpha P_{2x}(r_{FFL})+\beta P_{3x}(r_{FFL})=0 \\ {P}_{1z}(r_{FFL})+\alpha P_{2z}(r_{FFL})+\beta P_{3z}(r_{FFL})=0 \end{gathered}$$

This system can be solved linearly to determine α and β, which define the relative currents required in all three coils to position the FFL at any given (x,z) coordinate.

To determine the absolute current magnitudes, we introduce a third constraint: the desired FFL gradient magnitude G_desired. We first assume I₁ = 1 A and compute the total magnetic field using the previously determined ratios:

$$B_{tot}(x,z)=P_1(x,z)+\alpha P_2(x,z)+\beta P_3(x,z)$$

The local gradient magnitude at r_FFL is then computed numerically, yielding G₁, the gradient at unit current. Since the gradient scales linearly with current, the required scaling factor is $\text{λ}=G_{desired}/G_1$. The final currents are then:

$$\begin{gathered} I_1=\mathrm{\lambda } \\ {I}_2=\mathrm{\lambda }\alpha \\ {I}_3=\mathrm{\lambda }\beta \end{gathered}$$

This approach fully determines the coil currents required to generate an FFL at a given location with a specified gradient.

Although this formulation is specific to our three-coil configuration, it generalizes to systems with more than three selection coils. In such cases, additional degrees of freedom can be used to enforce further constraints, such as minimizing total power dissipation. While prior work has demonstrated x-position FFL control with only two coils Chinchilla et al. (2021); McDonough et al. (2024), such configurations lack independent control over FFL height and typically suffer from nonlinear FFL trajectories that degrade image quality. Using three or more coils ensures full control of the FFL position and gradient in the XZ plane and opens possibilities for further optimization.

2.5. Imaging Sequence

Imaging with our single-sided scanner is performed by electronically stepping the FFL across the FOV while measuring the SPIO signal at each discrete FFL position. This approach enables the acquisition of one-dimensional projections, which are subsequently reconstructed to generate tomographic images. While the general imaging procedure remains similar to that described in our previous work McDonough et al. (2024), enhancements have been made to the imaging sequence. Namely, the ability to scan larger FOVs, up to 8 cm (from 4 cm), and the ability to scan the FFL at a single height (17 mm), eliminating artifacts originating from inhomogeneous spatial encoding.

Each imaging sequence begins by defining a flat FFL trajectory in the XZ plane. The trajectory consists of evenly spaced FFL positions, typically 1 mm apart, covering a user-defined FOV that is symmetric about the scanner center. For example, a 4 cm FOV ranges from −20 mm to 20 mm. The user also specifies a desired magnetic field gradient that is held constant across all FFL positions to ensure compatibility with filtered backprojection (FBP) reconstruction.

The relative coil currents needed to generate the FFL at each (x,z) location are computed using the algorithm described in Sec.2.4. These currents are then applied to the selection coils using programmable DC power supplies: two (SM30-200, Delta Elektronika) for the symmetric near-surface selection coils, and another one (GENESYS+ TDK Lambda) for the central embedded selection coil controlled via Ethernet. For system stabilization, a 203 ms time delay is applied when the power supplies first form the FFL at the edge of the FOV. After the 203 ms wait time, the signal is measured for 70 ms, after which the FFL is stepped to the next position. At this new FFL position, there is a 47 ms wait time before data collection; this wait time is shorter since moving the FFL is less of a change to the system than forming it. After the 47 ms wait time, the signal is recorded for 70 ms. This process is repeated until the full trajectory is sampled.

To acquire 2D images, the sample is mechanically repositioned using the motorized stage shown in Fig. 1(a). For XY imaging, the sample is rotated in 9° increments using the stepper-motor-controlled rotational stage. After 20 projections over 180°, the resulting sinogram is reconstructed using filtered backprojection (FBP) with Hann filtering. For imaging applications requiring increased sensitivity, additional projections can be acquired by continuing rotation beyond 180°. For XZ imaging, no rotation is required. Instead, the phantom is translated vertically in discrete steps using the same sample stage. Projections are acquired at 31 z-positions with a step size of 0.5 mm. These projections do not require tomographic reconstruction and are stacked directly to form an image in the XZ plane.

The ability to define a variable FOV is a major improvement in system capability. There is a limit on the maximum achievable gradient at larger FOVs, since a uniform gradient must be maintained across the scan for accurate FBP reconstruction. The gradient strength can be increased at the price of reduced FOV, enabling focused imaging over a smaller region. This “zoomed-in” mode supports gradients up to 2 T/m for a 4 mm FOV. In this work, the smallest FOV used was 3 cm with a gradient of 1 T/m. By comparison, imaging over the full 8 cm FOV requires a maximum gradient of 0.4 T/m due to the reduced selection field strength at the edges.

Another major advancement in this work is the ability of the scanner to maintain a flat FFL trajectory at a constant height of 17 mm. In previous implementations, the FFL height varied with x-position, resulting in a curved trajectory and degradation artifacts near the edges of the FOV. Flat FFL control improves spatial uniformity and reduces reconstruction artifacts.

3. Results

In this section, we present imaging of various phantoms to demonstrate the enhanced performance of our scanner.

3.1. Imaging Sensitivity Study

To evaluate the imaging sensitivity of our MPI scanner, we imaged a series of dilution phantoms filled with varying concentrations of Synomag-D 70 nm. All scans in this study were performed with a 4 cm FOV, 0.6 T/m selection gradient, and 1 mm FFL step size, corresponding to 41 FFL positions per projection. Each full projection required 5 s, and the complete sinogram was acquired over 20 angles for a total imaging time of 100 s. The excitation coil was driven at I_RMS = 28.9 A, yielding an excitation field strength of 2.95 mT.

Dilution phantoms consisted of 0.5 mL microcentrifuge tubes containing 18 μl of SPIO solution at varying dilutions: 1:1, 1:4, 1:16, 1:32, 1:64, and 1:128, as seen in Fig. 5(b). Based on the specified Synomag-D concentration of 20 mg/ml, these dilutions correspond to total SPIO masses of 360, 90, 22.5, 11.3, 5.6, and 2.8 μg, respectively. Assuming the SPIO solution is composed entirely of magnetite, iron comprises 72% of the total mass, yielding a minimum visible iron content of 2.0 μg. This fraction was determined by calculating the iron content of magnetite (Fe₃O₄), where the atomic masses of Fe and O are 55.8 u and 16.0 u, respectively. The total molecular mass of Fe₃O₄ is therefore 3(55.8) + 4(16.0) = 231.4 u, of which 3(55.8) = 167.4 u corresponds to iron. Each phantom was positioned in the center of the imaging FOV, 17 mm above the surface of the scanner.

Figure 5.

Sensitivity characterization using a dilution series of SPIO-filled microtubes. (a) Cross-sectional MPI images of six dilution phantoms containing decreasing masses of SPIO (360 to 2.8 μg). (b) Photograph of the dilution series, showing the corresponding microtubes arranged by SPIO mass. (c) Signal plotted as a function of mass of solid SPIO content on a log-log scale. Measured values show a linear relationship between signal and iron content. The red line indicates the fit, and the blue line marks the noise floor.

All samples were scanned using the standard 100 s acquisition protocol, except the three lowest concentrations (11.3, 5.6, and 2.8 μg), which were each averaged over 15 repetitions to enhance SNR, resulting in a total scan time of 25 minutes. Background noise was characterized by imaging an empty FOV under the same acquisition conditions, with the maximum pixel value of this image corresponding to the noise level. MPI images and a corresponding plot of signal versus dilution are presented in Fig. 5(a,c). We observed a detectable signal (SNR>1) down to the 1:128 dilution, corresponding to 2.8 μg of SPIO, or 2.0 μg of iron.

3.2. Spatial Resolution Study

The spatial resolution of our scanner was assessed using a two-rod phantom with variable spacing. Each rod measured 20 mm in length and 1 mm in diameter and was filled with undiluted Synomag-D 70 nm SPIO at a concentration of 10 mg/mL. A photograph of the SPIO-filled rods is shown in Fig.6.

Figure 6.

Spatial resolution assessment using a two-rod phantom filled with undiluted Synomag-D 70 nm (10 mg/mL). (Top row) Reconstructed MPI images for separations of 2, 4, 5, 6.5, and 8 mm. (Bottom row) Corresponding horizontal cross-section taken across the center of each image. Rods become clearly resolvable at 4 mm separation, establishing the system’s spatial resolution. (Right) Photograph of the phantom with 1 mm diameter, 20 mm long rods.

Imaging was performed over a 3 cm FOV with a 0.5 mm FFL step size, corresponding to 61 positions per projection. The reduced FOV permitted the application of a stronger selection gradient of 1 T/m. Each projection required 7.34 s, resulting in a total acquisition time of 147 s for 20 angles. The excitation coil was driven at I_RMS = 20 A, producing an excitation field strength of 2.0 mT.

Five inter-rod spacings were tested: 2, 4, 5, 6.5, and 8 mm (side to side). As shown in Fig.6, the rods were clearly distinguished at 4 mm separation, establishing a conservative spatial resolution of 4 mm.

3.3. Flat Trajectory Study

To evaluate the effectiveness of the FFL control algorithm, we imaged a triple-rod phantom using both corrected and uncorrected FFL trajectories, in both the XZ and XY imaging planes. The phantom consisted of three rods (1 mm diameter, lengths 25 mm) filled with undiluted Synomag-D 70 nm SPIO at a concentration of 10 mg/mL. The rods were aligned at the same height within a 3D printed holder and spaced 20 mm center-to-center. Photographs of the phantom and holder are shown in Fig.7(d,h).

Figure 7.

MPI images of a triple rod phantom acquired with and without the FFL trajectory correction algorithm. (a-c) XZ plane images of the triple rod phantom: (a) imaging with corrected FFL trajectory, (b,c) Imaging without FFL height correction leading to a curved trajectory. Red circles denote the locations of the SPIO-filled rods. The white curves indicate the slices at which the corresponding XY images were acquired. (e-g) XY plane images acquired at the heights marked in (a-c), respectively: (e) corrected trajectory shows accurate geometry of rods, (f) curved trajectory causes the central rod to appear shortened and the side rods to be missing, (g) imaging with the curved trajectory at a different height causes the side rods to appear lengthened and curved, while the central rod is missing. (d) Photograph of the triple rod phantom in a holder with 20 mm spacing between rods. (h) Photograph of the individual SPIO-filled rods used in the phantom.

For all imaging conditions, we used a 6 cm FOV with a 1 mm FFL step size, corresponding to 61 positions per projection and a 7.34 s acquisition time per projection. For XY imaging, projections were acquired at 20 different angles, resulting in a total scan time of 147 s. For XZ imaging, the phantom was translated vertically in 0.5 mm increments across 31 different z-positions, corresponding to a 228 s scan time. A constant gradient of 0.6 T/m and an excitation strength of 2.0 mT (I_RMS = 20 A) were used for both scans. Reconstructed MPI images of the phantom are shown in Fig.7(a–c,e–g).

With the corrected FFL trajectory, all three rods were clearly resolved with accurate lengths, spacing, and positions in both the XY and XZ planes Fig.7(a,e). Alternatively, in the previous approach, the three rods were placed along the arc trajectory to be imaged simultaneously by the scanner McDonough et al. (2022a). In contrast, imaging of linearly arranged objects with the original curved trajectory produced distorted results. The XZ image remained spatially accurate in rod positioning, but the image was non-rectangular due to the arc-shaped trajectory of the FFL, Fig.7(b,c). The white curves in Fig.7(b,c) indicate the imaging heights used for the corresponding XY images. Imaging at the sample height where the FFL intersects the central rod resulted in a shortened central rod and omission of the side rods Fig.7(f). Conversely, imaging at the height intersecting the side rods resulted in an image missing the central rod, with the outer rods appearing lengthened and warped Fig.7(g).

3.4. FOV Study

To demonstrate the extended imaging FOV enabled by these upgrades, we imaged two phantoms entirely in the FOV: a planar spiral phantom and an anatomically shaped mouse phantom in the coronal plane. The mouse phantom includes compartments representing major organs and two “tumor” volumes, each filled with a 1:2 dilution of Vivotrax (Magnetic Insight, USA) SPIO. The spiral phantom was constructed from 1 mm inner diameter tubing embedded in a flat 3D-printed former, filled with undiluted (10 mg/mL) Synomag-D 70 nm SPIO, forming three full turns with an initial radius of 35 mm and a radial taper of 12 mm per turn. Photographs of both phantoms are shown in Fig.8.

Figure 8.

(a) Photograph of the spiral phantom constructed from 1 mm inner-diameter tubing embedded in a 3D-printed former.(b–c) Photographs of the anatomic mouse phantom with labeled organ and tumor compartments filled with Vivotrax SPIO. (d) Reconstructed MPI image of the spiral phantom. (e) Reconstructed MPI image of the mouse phantom acquired with an 8 cm FOV (gradient 0.4 T/m). (f) Reconstructed MPI image of the same phantom acquired with a 6 cm FOV (gradient 0.6 T/m).

Both phantoms were imaged using an 8 cm FOV with a 1 mm FFL step size, resulting in 81 discrete FFL positions per projection. A gradient strength of 0.4 T/m and an excitation field of 1.8 mT (I_RMS = 18 A) were used for all scans. Each projection required 9.68 s, corresponding to a total imaging time of 194 s (20 angles). Reconstructed MPI images of both phantoms are shown in Fig.8.

Signal fading is observed near the edges of the 8 cm FOV for both phantoms, as shown in Fig. 8(d,e), due to the receive coil’s sensitivity profile, which is weaker further from the center of the FOV. To recover imaging homogeneity and improve spatial resolution, the mouse phantom was also imaged using a reduced 6 cm FOV, which allowed the gradient to be increased to 0.6 T/m. In this experiment, all organ compartments were filled except the brain and upper tumor volumes. The scan was performed with a 0.5 mm FFL step size, yielding 61 discrete FFL positions per projection. The excitation field was maintained at 1.8 mT (I_RMS = 18 A), and each projection required 7.34 s, resulting in a total imaging time of 147 s (20 angles). The reconstructed image for this configuration is shown in Fig. 8(d).

4. Discussion

In this work, we presented the next generation of the FFL MPI scanner featuring a hybrid surface-volume excitation coil, providing a semi-homogeneous magnetic field over an 8 cm FOV, addressing a challenge inherent to single-sided systems. In addition to hardware upgrades, a new algorithm was implemented for electronic FFL trajectory control, enabling flat scans across extended FOVs. These hardware and software enhancements improve the scanner’s performance (see table 1) and clinical viability by mitigating field inhomogeneity and enabling linear spatial encoding.

Table 1.

Comparison of metrics of MPI scanners.

	Original Single-Sided FFL Scanner1	Enhanced Singled-Sided FFL Scanner	Other Human Scanners2
Drive Amplitude (mT)	0.9	4.0	6–7
Gradient (T/m)	0.58	1.0	0.2–1.0
Sensitivity (μg Fe)	6.0*	2.0	1.0–19
Spatial Resolution (horizontal, mm)	6.5**	< 4.0	–
Maximum FOV (cm2)	4.0 × 4.0	8.0 × 8.0	–

^*Theoretically estimated sensitivity.

^**Obtained with PSF deconvolution.

¹ McDonough et al. (2024)

²Graeser et al. (2019); Mattingly et al. (2025); Vogel et al. (2023); Wei et al. (2024)

The sensitivity study demonstrated reliable detection of a 1:128 dilution, corresponding to 2.8 μg (2.0 μg of iron). Extrapolating the linear fit to the data (see Fig.5) to the background level (SNR=1) implies that we should be able to reach a sensitivity of 1.0 μg of iron content, providing a lowered noise level. This meets a clinically relevant sensitivity threshold of 4.8 μg for tumor imaging, based on a 400 mg iron dose, 1% ID/g tumor uptake, and 1.2 mg tumor mass, bringing us closer to our desired application of breast tumor detection. In this study, the excitation field was limited to 2.95 mT due to heating of the excitation coil, resulting in current instabilities. Higher excitation fields, and therefore improved sensitivity, could be achieved through more efficient cooling of the excitation coil. This could be implemented by submerging the coil in a thermal bath or constructing it from hollow conductors with internal water flow.

The spatial resolution study demonstrated a 4 mm resolution, defined by the minimum distinguishable separation between two SPIO-filled rods. This resolution was achieved using FBP without additional enhancements from iterative reconstruction or deconvolution. The demonstrated 4 mm resolution is clinically relevant for breast cancer applications, which are a primary focus of this system. The median tumor size detected by current screening modalities is 7.5 mm Michaelson et al. (2003), and most early-stage tumors fall within the 5-10 mm range. Moreover, MPI’s zero signal from tissue allows for the detection of lesions smaller than the nominal resolution, provided adequate sensitivity. Clinically, breast-conserving surgery typically recommends a minimum margin of 2 mm Morrow et al. (2016), suggesting that a spatial resolution of 4 mm is sufficient to support margin assessment. These results demonstrate the scanner’s suitable spatial resolution for breast imaging and its potential utility for surgical planning and guidance.

The accuracy and utility of the FFL trajectory control algorithm were demonstrated in the flat trajectory study, where a triple-rod phantom was imaged in both the XY and XZ planes with and without correction. Without trajectory correction, the reconstructed images exhibited significant artifacts, including distorted geometry and missing rods. These artifacts result from the nonlinear FFL trajectory, which causes inconsistent spatial encoding across the FOV. In contrast, with the correction algorithm applied, all rods were clearly visualized in both imaging planes with accurate lengths, spacing, and positions. These results confirm that the trajectory correction algorithm performs as intended and extends the usable imaging FOV beyond the previous 4 cm limit.

The extended 8 cm FOV was validated using two phantoms: a spiral phantom and a mouse phantom with anatomical volumes. Both were successfully imaged over the full FOV, demonstrating the system’s ability to accommodate extended geometries and larger subjects. An 8 cm FOV represents a substantial advance for single-sided MPI scanners and is larger than that of leading commercial preclinical systems. For example, the Magnetic Insight Momentum offers a 6 cm lateral FOV, while the Bruker MPI scanner provides a 5 cm lateral FOV.

While the 8 cm FOV was successfully imaged, signal attenuation was observed at the FOV edges (Fig.8). This attenuation is attributed to the sensitivity profile of the 7 cm diameter receive coil, which loses sensitivity towards the edges of the 8 cm FOV. Increasing the size of the receive coil would enhance sensitivity homogeneity across the 8 cm FOV. However, with our current receive gradiometer design McDonough et al. (2022b), scaling the receive coil necessitates scaling the gradiometer, as cancellation relies on an identical receive gradiometer pair. Due to geometric constraints imposed by the surface-volume excitation coil (Fig.2), the receive gradiometer pair cannot be scaled beyond 7 cm. Consequently, for a symmetric receive gradiometer system, 8 cm represents the maximum FOV.

Homogeneity can be recovered by reducing the FOV, as shown in the mouse phantom study. Restricting the FOV to 6 cm confines the imaging region to the semi-homogeneous zone of the receive coil’s sensitivity profile, resulting in a more uniform image. This “zoomed” FOV also permits operation at a higher gradient strength of 0.6 T/m, which restores resolvability between organs such as the kidneys (Fig. 8d). Comparing the full-FOV image (Fig. 8e) to the zoomed image (Fig. 8f) demonstrates that both image uniformity and organ resolvability are possible when the FOV is zoomed to this region.

Future work will aim to extend homogeneity across the entire 8 cm FOV through both hardware and software improvements. On the hardware side, alternative receive–gradiometer designs could be explored to achieve a more homogeneous sensitivity profile across larger FOVs. On the software side, scanner performance could be further enhanced through an image reconstruction upgrade. In this work, reconstruction was accomplished with FBP, without post-processing. Replacing FBP with model-based iterative reconstruction that incorporates a measured or simulated system function should raise both sensitivity and spatial resolution Chacon-Caldera et al. (2021), additionally, this could correct for the current limited sensitivity profile of the receive coil, eliminating the inhomogeneity at the edges of the large FOV (8 cm). Alternatively, post-processing techniques, such as deconvolution or machine-learning techniques, could further enhance scanner performance as well. Standard deconvolution techniques are challenging in single-sided systems because they assume spatially invariant point-spread functions, which is a challenging assumption for scanners with inhomogeneous magnetic fields. Although single-sided scanners to this date feature inhomogeneity in imaging, this may be mitigated in the future.

5. Conclusions

In this study, we presented a next-generation single-sided FFL MPI scanner featuring a hybrid surface-volume excitation coil and a new algorithm for FFL trajectory control. These upgrades enabled significant improvements in imaging homogeneity, sensitivity, spatial resolution, and imaging FOV. We demonstrated a clinically relevant SPIO detection limit of 2.0 μg of iron and a spatial resolution of 4 mm without the use of iterative reconstruction or post-processing, comparable to or exceeding the performance of other human-compatible scanners. The system achieved a FOV of 8 cm, exceeding that of leading commercial preclinical scanners, and successfully imaged entire extended and anatomically realistic phantoms. Flat FFL trajectories were verified through multiplane imaging of a triple-rod phantom, confirming precise electronic control over FFL positioning. These results demonstrate a substantial increase in imaging capability of single-sided MPI, bringing the modality closer to clinical relevance.