MPI visualizes SPIO tracers by direct induction, so the MPI signal scales exactly linearly with the number of SPIOs in the imaging volume, meaning MPI is a “hotspot” imaging technique without background signal, similar to nuclear medicine techniques. For stem cell tracking , MPI provides quantitative cell counts anywhere in the body for up to months in the same animal [13]. Unlike optical imaging, ultrasound, X-ray, and other imaging methods, the MPI magnetic signal suffers zero attenuation with depth and there is no MPI signal from background tissues. The MPI signal also has excellent mass sensitivity because MPI detects the intense electronic magnetization of an SPIO, which is 22 million times more intense than the nuclear paramagnetism of water in 7T MRI. Hence, our prototype MPI scanner can detect even 2 nanograms of SPIOs within a single voxel. Theoretically, this translates to roughly 200 nanomolar concentration of iron, or to 200 cells with in vivo scans in state-of-the-art MPI scanners. Hence, the contrast and sensitivity of MPI rival nuclear medicine techniques like PET and SPECT without concerns about radiation dose or “half-life” constraints. Indeed, a few SPIO formulations have previously been approved by EU and FDA as diagnostic agents. MPI uses no ionizing radiation, and it can operate within the human magnetic safety limits so human translation will soon be feasible. Current disadvantages of MPI include relatively poor spatial resolution (roughly 1 mm in a mouse), and the lack of anatomical information implying that it cannot be used as a single imaging modality alone. Recently, a commercial murine MPI scanner was released in Germany (Bruker), and a preclinical scanner is in development at Magnetic Insight (California). A start-up in Seattle (LodeSpin Labs ) has been formed around the synthesis of MPI-tailored SPIOs.

The physics governing MPI is completely different from MRI, and it is not possible to perform MPI with an MRI scanner or vice versa. An overview of the physics, hardware, and reconstruction theory of MPI can be found in references [2–4, 11, 12, 14–17, 21, 48]. The SPIO tracers used in MPI obey steady-state Langevin physics as shown in Fig. 2. They achieve steady-state alignment shown here with applied fields in tens of microseconds. MPI scanners employ a strong gradient field to force all SPIOs outside of the gradient field origin, called the field-free point (FFP) , to be saturated. Saturated SPIOs induce no signal in the MPI inductive receiver coil because they are static.

All MPI scanners today scan the FFP across the FOV at roughly 25 kHz to create an MP image, as illustrated in Fig. 3. The SPIOs flip their orientation when the FFP sweeps past their location. The flip of even 2 nanograms of SPIOs in a voxel induces a detectable signal in a receiver coil. Importantly, the induced signal is linearly proportional to the SPIO mass at each location in space. Hence, the time axis of the MPI induction signal maps directly to spatial location in the SPIO image. This is the intuition behind the x-space MPI reconstruction algorithm [11, 12, 14–17, 48]. Multidimensional MPI is a simple extension of the 1D imaging process. Movement of the FFP in 3D is implemented by dynamically driving three spatially homogeneous “slow shift field” coils, one aligned with each direction (x, y, z). This effectively scans the FFP, or field-free line (FFL) , in two or three dimensions, as shown in Fig. 4.

Unfortunately, all MPI scanners must deal with a crucial technical challenge. The received MPI signal occurs at the same time as the drive field, which can be orders of magnitude stronger in amplitude. Hence, the receive coil picks up a massive interference from the time-varying drive fields. Fortunately, this direct feed-through interference is localized (mostly) at the drive frequency, typically 25 kHz, by meticulous design of the transmit power electronics to ensure a low total harmonic distortion at 10–100 ppm. Hence, while the MPI signal is not contaminated by direct feed-through at higher harmonics, the signal at the fundamental frequency is swamped with direct feed-through interference. As a result, all MPI scanners reject the direct feed-through using a sharp passive filter, with more than 30,000-fold suppression. Unfortunately, this filter also removes the first harmonic content of the SPIO signal. The MPI image reconstruction challenge is to reconstruct a high-quality SPIO image using only the SPIO signal derived from higher harmonics of the 25 kHz drive frequency.

Both the system matrix reconstruction approach, advanced by Philips, and the x-space reconstruction approach, developed by UC Berkeley, have shown that the lost first harmonic information corresponded precisely to the average (baseline or DC) signal in a partial imaging volume [4, 11]. Hence, it is possible to design a robust image reconstruction algorithm that restores the lost DC information using excitation trajectories with overlapping partial imaging volumes, or field of views (pFOV). These are then stitched together using a fast continuity algorithm [11], which is demonstrated in Fig. 5. Importantly, Lu et al. and Zheng et al. proved experimentally and theoretically that MPI is a linear and shift invariant imaging system after the lost baseline information is restored [11, 13]. Several studies have also proven that the fundamental 1D point spread function (PSF) of MPI is simply a scaled version of the derivative of the SPIO’s Langevin function, as shown in Fig. 2 [14, 15, 17]. Below we summarize the fundamental properties of MP images: contrast, sensitivity, and resolution.

Unlike in optical or ultrasound imaging, biological tissue does not attenuate the MPI signal. There is also no signal from living tissue in MPI as demonstrated experimentally in Fig. 6. This makes MPI uniquely suitable for linearly quantitative and radiation-free cell tracking at any depth in vivo.

As shown in Fig. 7, the MPI image intensity signal scales linearly with the mass of SPIO particles at each pixel due to the linearity and shift invariance properties of MPI. Therefore, the MPI signal is perfectly linear and quantitative with cell count. MPI directly detects the SPIO electronic magnetization, which is 22 million times more intense than the nuclear paramagnetism imaged in 7 T high-field MRI. Direct detection of SPIO tracers enables “hotspot imaging” in MPI as opposed to the “negative contrast” when the same SPIOs are used in T₂*-MRI. We have experimentally demonstrated a linear detection limit of 200 labeled cells in vitro [13], with potential for significant further improvement. This linear sensitivity is superior to all other imaging modalities, the detection limits of which are compared in [1].

Importantly, although SPIOs respond nonlinearly to the applied magnetic field, the MPI signal induced from the rotating SPIOs in the receiver coil is perfectly linear with respect to the mass of SPIOs. Hence, there is absolutely no saturation of the MPI signal at higher cell counts. The details of the linearity and shift invariance of MPI are shown in [11, 14].

The fundamental spatial resolution of MPI depends only on the saturation characteristics of the SPIO particles and the strength of the magnetic gradient. Doubling the gradient strength improves resolution by twofold in every dimension. Currently available SPIOs saturate at roughly plus or minus 3 mT, which translates to a spatial resolution of roughly 1 mm in a 6T/m gradient field. While this resolution is comparable to preclinical nuclear medicine, it is not yet competitive with CT or MRI small-animal imaging. To improve MPI’s spatial resolution we and others have studied SPIOs with varying core diameters, since Langevin physics predicts a cubic resolution improvement with increasing SPIO core size. This has produced exciting results effectively reducing the FWHM resolution by a factor of two over Resovist [8]. Unfortunately, researchers have noted that improvements beyond roughly 25 nm core diameter seem to be limited, perhaps due to relaxation-induced blurring [18, 19]. This area remains a major active research area for tailoring nanoparticles to improve MPI performance.

All of the imaging data shown in this chapter are obtained from the two scanners shown in Fig. 8. The top shows our 2.4,T/m FFL scanner , which allowed us to perform fast projection imaging as well as 3D projection reconstruction imaging, which is similar in scanning and reconstruction to X-ray computed tomography [48, 49]. Because we synthesize the 3D images from N projections, this volumetric imaging method improves image SNR by the square root of N over standard FFP imaging. We employed both the x-space reconstruction algorithm and the filtered back projection algorithm to reconstruct these images. The bottom scanner is our 7T/m FFP MPI scanner, which has produced some of the highest spatial resolution MP images reported thus far.