1 INTRODUCTION

Imaging techniques are indispensable for the diagnosis and treatment of ischemic stroke and other neurological disorders. In addition to ruling out hemorrhages and other neurological diseases mimicking acute ischemic stroke symptoms, a major objective of imaging patients with an acute ischemic stroke is determining the volume, location of the infarct core, and the surrounding ischemic penumbra (Baird et al., 1997). Physiologically, a cerebral artery occlusion does not lead to an immediate, irreversible injury of corresponding brain regions due to a reduced blood supply by collateral vessels such as the circle of Willis or leptomeningeal collaterals. Following these physiological considerations, the infarct core is defined as the irreversibly damaged tissue. In contrast, the hypoperfused brain tissue within the penumbra is dysfunctional and at an increased risk of necrosis but still salvageable and represents the target for any treatment approach. Patients with a small core and a large penumbra are likely to benefit most from any intervention (e.g., thrombectomy and thrombolysis). Thus, the ratio of the stroke core and penumbra volume is often used for treatment selection. Unfortunately, the infarct core expands into the penumbra over time in a dynamic process. This observation has led to the concept of “time is brain” for treating acute ischemic stroke patients, postulating that a faster recanalization of the occluded artery leads to more preserved tissue and an improved clinical outcome. So, any image-based diagnosis needs to be quick (Saver, 2006).

PET studies revealed that salvageable tissue exists for up to 24 h. Consistent with these findings, more and more studies show that patients might benefit from thrombectomy even after more than 12 h. Therefore, treatment decision-making in acute ischemic stroke patients is increasingly based on imaging information in addition to clinical criteria, including the time window. Stroke therapy is becoming more and more personalized medicine. Thus, more precise imaging is needed to identify with greater accuracy those patients who benefit from therapy in an extended time window. However, the low temporal resolution, greater than 1 s, in standard CT and MRI perfusion imaging can have a significant effect on the quantified infarct core and penumbra segmentations, depending on the quantification method. This variation can deprive patients of optimal stroke treatment. Magnetic Particle Imaging (MPI) is a tomographic imaging technique with a high temporal resolution of up to 46 frames per second in 3D(+t) datasets (or even more than 1000 frames per second in 2D(+t) datasets). MPI may enable faster, more reliable, and more accurate identification of the infarct core and penumbra and subsequently improve prediction of which patients will benefit from stroke treatment. This review will highlight the first preclinical stroke imaging studies and the recent progress in the development of MPI scanners for humans.

2 MAGNETIC PARTICLE IMAGING

The physical principle of MPI relies on the non-linear shape of the particle magnetization curve (Gleich & Weizenecker, 2005). When the particles are excited by an oscillating magnetic field, the magnetization behavior acts as a non-linear mixer of all applied fields (Rahmer et al., 2009). If the magnetization response is recorded using inductive sensors, the induced signal shows higher harmonics and, in the case of multidimensional excitation, mixing of all excitation frequencies. As the shape of the excitation field is typically purely sinusoidal, the presence of these higher frequency components already encodes the presence of tracer material within the field of view (FOV). To ensure a spatial encoding, a linear gradient, the so-called selection field, is applied featuring a low field region in its center. This region can have the shape of a line (referred to as field-free line, FFL) or a point (referred to as field-free point, FFP) depending on the field generator topology (Weizenecker et al., 2008). The superposition of the two fields results in a unique field vector sequence at every point in space. In consequence, the magnetization time sequence in this spatial position is unique as well. Under the assumption that the particle response is linearly dependent on its concentration, which holds true for most tracer systems at clinically relevant concentrations, the imaging problem can be described as a linear system of equations (Grüttner et al., 2013; Lu et al., 2013), which can be solved using iterative linear solvers (Knopp et al., 2010). The basic principle of signal generation, spatial encoding, and reconstruction is shown in Figure 1.

Fundamental principles of MPI. MPI images the distribution of nanoparticles within a defined FOV. In the example, a 1D sinusoidal excitation field is applied (a). In superposition with a gradient field seen on the left side, the excitation field moves the low field region over space in time, defining the field of view (FOV). Tracer distributions within the FOV experience a changing magnetization causing signals in receive coils nearby. As the selection field is unique in strength and orientation throughout the FOV, the field sequence and, therefore, the signal response is also unique. This can be directly seen in the signal spectrum on the right side, which acts as a local fingerprint encoding a specific position in space (b). The image reconstruction shown on the bottom uses this fingerprint and determines the spatial distribution by solving a linear system of equations

The imaging performance is mainly determined by the instrumentation and the magnetization properties of the tracers. The spatial resolution of the MPI is driven by a combination of the slope of the dynamic magnetization curve on the one hand and the gradient strength of the encoding field on the other hand (Gleich & Weizenecker, 2005). Another parameter affected by the tracer and the instrumentation is the sensitivity (Knopp et al., 2011). Similar to the spatial resolution, steep magnetization curves provide good sensitivity. The sensitivity is influenced mainly by the coil and amplifier noise on the instrumentation side, as patient noise dominance is hard to achieve below 500 kHz (Schmale et al., 2010). The temporal resolution is mainly dependent on the instrumentation, namely on the selection of the drive field frequencies. In multidimensional excitation systems, these frequencies need to be chosen in a way that they are dividers of a common base frequency (Knopp et al., 2008). The greatest common divisor of those frequencies defines the repetition frequency and, therefore, the frame rate. However, when designing the system, the chosen frequencies also impact the signal induction, the dynamic magnetization curve, and the demands on the hardware. Therefore, a compromise between speed, hardware efforts, and particle dynamics needs to be found. An overview of the history of different scanner developments can be seen in Figure 2.

History of MPI. After invention in 2001, Bernhard Gleich and Jürgen Weizenecker presented the first small animal MPI scanner (Gleich & Weizenecker, 2005). After that, several groups worked on different field geometries (Draack et al., 2021; Goodwill et al., 2012; Sattel et al., 2009; Vogel, Rückert, et al., 2014), upscaling (Gleich et al., 2010), and multimodal solutions (Franke et al., 2016; Vogel, Lother, et al., 2014; Vogel, Markert, et al., 2019). In 2014 and 2016, the first two commercial small animal imaging units were introduced by Bruker BioSpin and magnetic insight. The first two scanners suitable for clinical use were presented in 2015 by the Philips research group (Rahmer et al., 2018) and 2019 by the University Medical Center Hamburg (Graeser et al., 2019)

One of the main advantages of MPI is the capability to image fast dynamic processes within the body. In preclinical systems, MPI can reach a frame rate of up to 1050 frames per second (Vogel et al., 2020), which allows MPI to capture even fast dynamic movements like blood flow within the aortic arch. As the aortic arch is the region of the most rapid velocities within the human body, MPI proves to be capable of imaging any important physiological dynamic without the need for triggering techniques or multiple tracer injections. The imaging speed is mostly determined by the chosen excitation frequencies as well as the trajectory density. It is independent of the bore size, which makes such high speeds possible in human-scale systems. If a specific measurement protocol does not provide enough imaging speed, a preprocessing step before image reconstruction allows reconstructing processes that are repetitive in nature using reordering of the raw data. Thus, artifact-free imaging without triggering is even possible for low imaging speeds (Gdaniec et al., 2020). The spatial resolution can reach down to 330 μm (Vogel, Markert, et al., 2019; Vogel, Rückert, et al., 2019), while most systems have resolutions in the range of 1–2 mm (T. Knopp et al., 2017). As a gradient strength of about 2–3 T/m/μ0, which is used in preclinical scanners, is also feasible for human-scale scanners, a similar resolution is possible in human applications (Rahmer et al., 2018). The sensitivity of MPI systems currently reaches down to about 900 pg iron content using optimized receive coils and electronics commercially available tracer material (Graeser et al., 2017, 2020). This allows long-time continuous imaging over several days without iron overdose. The high sensitivity also enables cell tracking for a very low number of cells, probably down to 1–10 cells (Graeser et al., 2020; Zheng et al., 2015), providing more possibilities for stem cell research.

As described above, the imaging process relies on the dynamic magnetization curve of the tracer material. This dynamic process relies on several parameters, including the physical parameters of the tracer like the base material, crystal structure, shape, and the coating size and material (Rauwerdink et al., 2009; Rauwerdink & Weaver, 2010a, 2010b; Weaver et al., 2009); (Graeser, Bente, & Buzug, 2015; Graeser, Bente, Neumann, & Buzug, 2015; Viereck et al., 2017). Additionally, the local microenvironment of the tracer influences the magnetization behavior, for example, the temperature of the medium or the viscosity. In multicontrast (or multicolor) MPI, this dependency is exploited to use the tracer as a microsensor and encode these physiological parameters within the image domain (Haegele, Panagiotopoulos, et al., 2016; Möddel et al., 2018, 2020; Rahmer et al., 2015; Shasha et al., 2019; Stehning et al., 2016; Szwargulski et al., 2019; Szwargulski et al., 2020; Utkur et al., 2019). Figure 3 outlines different possibilities of multicontrast MPI.

Application fields of multicontrast MPI. Multicontrast MPI (top left) allows the determination of additional parameters like the viscosity (Möddel et al., 2018) (top right) or the temperature (Salamon et al., 2020) (bottom middle). Alternatively, multicontrast MPI can also be used to image multiple tracers simultaneously (Szwargulski et al., 2018) (bottom left) or even to determine the size of multiple tracer systems (Shasha et al., 2019)

Most of the scientific research in MPI currently focuses on the preclinical side as only a few imaging systems with a human-sized bore exist. When scaling up MPI systems, new challenges arise both on the physiological and technological sides. The main technical challenges are a quadratic rise of the power consumption for all field generators and a rising voltage of the drive field coils caused by higher currents and larger inductance due to larger coil diameters.

Considering the imaging performance of human-scale systems, a larger diameter results in weaker coupling of the receiver coils with the tracer magnetization. This leads to a lower sensitivity which roughly scales with the diameter if a coil noise dominance is assumed. Due to the increased tracer dosage compared with rodents, this effect is compensated for most applications. Recent developments using optimized receive coils and receive electronics showed rapid improvement of the achieved sensitivity, which will also impact the sensitivity of human-sized systems (Graeser et al., 2017, 2020; Zheng et al., 2015).

The third impact of upscaling is physiological limits, namely the peripheral nerve stimulation (PNS) and the tissue's specific absorption rate (SAR). The oscillating magnetic fields induce eddy currents in human tissue, which can excite peripheral nerves at low drive field frequencies. At higher frequencies, the SAR dictates a lower limit than PNS (Saritas et al., 2015, 2013). A PNS and SAR study by Schmale et al. (Saritas et al., 2013; Schmale et al., 2015) showed that for human torso scale systems, the PNS limit is roughly around 3 mT. As these scales down the FOV, low-frequency focus fields are introduced, which consequently cover a larger region with patches of these smaller FOVs. The drive field frequency can be shifted to a higher region until SAR limits become dominant to recover the high temporal resolution.

In summary, the upscaling faces three opposing requirements that determine the complexity of the imager. The first is imaging performance parameters like the temporal and spatial resolution and the sensitivity. Different medical applications need a distinct focus on these imaging parameters, while a “one fits all” machine needs to cover all parameters at once. The second requirement is physiological and electrical limits that need to be met to ensure the safety of the patient and the operator. These cannot be overcome and also have to apply for medical regulation standards. The third one is the technological complexity of the system. As described above, the “one fits all” machine may be advantageous as it provides the physician a variety of imaging possibilities and raises the hardware side's effort to its limits. Besides rising costs, this leads to a spatially fixed system with high room infrastructure and space demands.

Up to now, two human-sized MPI systems were presented, one capable of torso imaging and the other intended for brain imaging. The first one was presented by Rahmer et al., 2018.

This system was aimed as a “one fits all” machine featuring a bore fitting a human torso. It provided a selection field of 2.8 T/m/μ0 resulting in an expected resolution of around 1 mm (Bontus et al., 2015). The drive field frequency was chosen to be 150 kHz with 3 mT/μ0 field strength resulting in a frame rate of 568 volumes per second (Rahmer et al., 2018). To cover a large FOV, the system provided focus fields of 8 mT, which makes the system the most flexible until today. On the other hand, the system had comparable infrastructural needs as an MRI tomograph being permanently installed and a large infrastructure room.

The second human-scale MPI system is a brain scanner developed to perform surveillance imaging directly on the stroke or intensive care unit (see Figure 4; Graeser et al., 2019). The aim of this head scanner is a specialized system that provides the imaging parameters needed for monitoring applications but reduces the technical effort to a minimum using a low-field approach. The system currently provides a selection field of only 0.25 T/m/μ0 with a drive field frequency of 25 kHz and 6 mT/μ0 excitation field strength. With these parameters, the system is capable of reaching a sensitivity limit of 263 pmol/LFe with a frame rate of 2 Hz and a spatial resolution of 5 mm. The clear advantage of the low-field approach is its small footprint of roughly 1 m2 and the capability of working on a standard 230 V power plug. In addition, the system is self-shielded and does not need any further infrastructure as cooling or radiofrequency shields.

Overview of the neurological MPI scanner. The selection field generator (SeFo) is placed under a copper shield. It can be seen in the cutout on the CAD model of the scanner. The working scanner can be seen in the scanner photo. In a static experiment, a human-sized phantom was filled with an effective iron concentration of 965 ng/mL Fe. A small region on the side was left empty to simulate a stroke of 42 mL. The stroke region is visible on the right side of the reconstruction. The capability to generate perfusion maps is shown on the right side, where a phantom filled with glass spheres was perfused with water. With a bolus injection of tracer, the perfusion parameters were calculated and visualized in parameter maps. Reprinted with permission from Graeser et al. (2019)

Another, yet not finished, demonstrator is built at the St. Martinos Center for Biomedical Imaging. The aim is to develop a human-sized functional magnetic particle imaging system and exploit the potential 40 times higher contrast to noise ratio of MPI (Cooley et al., 2018; Mason et al., 2017).

As a tracer-based imaging modality, the reconstructed images provide no anatomical reference and are, therefore, sometimes hard to judge. To overcome this challenge, several systems have been developed on a preclinical scale to combine MPI with anatomical imaging modalities like MRI or CT (Franke et al., 2016, 2020; Vogel, Lother, et al., 2014; Vogel, Markert, et al., 2019; Vogel, Rückert, et al., 2014). Due to the fast imaging speed, MPI can also be combined with optical coherence tomography (OCT) to provide intravenous tracer tracking and use this information to incorporate prior knowledge for the OCT image reconstruction (Griese et al., 2019; Latus et al., 2019).

The development of human-scale MPI systems is still in its beginnings. One research field is the optimization of the field generator, especially the section field generator. The selection field strength can still be improved with power optimization techniques and effective iron field guide design. As this strength is directly coupled to the spatial image resolution, the currently achieved image resolution does not represent the physical limit for MPI. For large volumes, the improvement in tracer performance can be translated into larger scan volumes by reducing the gradient. As different groups contributed to scientific achievements over the past years with different technological foci, the true potential of MPI lies in combining these features into one machine.

3 TRACER MATERIAL FOR MAGNETIC PARTICLE IMAGING

As a tracer-based imaging modality, the entire imaging performance inherently depends on the scanner instrumentation and the tracer characteristics. Therefore, intensive research on the optimal MPI tracer has been carried out in the past decade, leading to various approaches for optimizing image quality.

The research can be divided into two areas: optimizing the tracer core to improve the MPI performance regarding its spatial resolution and sensitivity and functionalizing the tracer surface to target specific medical applications.

Tracers in MPI consist of a magnetic core, which defines the magnetic properties, and a non-magnetic coating with two main functions: First, it prevents agglomeration, and, second, it defines its physiological properties. The core is usually made of magnetite (Fe3O4) with a high magnetic moment of 4.1μB, where, μB is the Bohr magneton (Petrov & Ustinov, 2010).

In the early years of MPI, ferucarbotran (Resovist), a clinically approved tracer with a magnetite core, was commonly used. Later, other commercially available tracers like Perimag (and VivoTrax) based on ferucarbotran and maghemite nanoflowers (γ-Fe2O3) like Synomag showed their potential in MPI. Figure 5 compares Resovist (batch: 81049S), Perimag (lot: 05617102-05), and Synomag (lot: 08219104-03) in a magnetic particle spectroscopy (MPS) experiment, which is an established method for determining the MPI imaging performance with a 1D field excitation without spatial encoding (Biederer et al., 2009). The three samples contained an iron amount of 56 μg in 15 μL resulting in an iron concentration of 66.6 mmol/L. Using a custom-made and calibrated MPS, they were exposed to a magnetic field amplitude of 20 mT/μ0 at 26 kHz. The measurements revealed an average amplification factor of 1.5 between Perimag compared with Resovist and 2.2 between Synomag and Resovist, which directly translates into an increase in sensitivity when using the tracer, providing a higher MPI signal.

Comparison of MPI tracers. The figure shows MPS measurements comparing Resovist (black), Perimag (blue), and Synomag (orange) with an iron amount of 56 μg in a 15 μL (66.6 mmol/L) solution. The derivative of the magnetic moment is shown for one excitation period in (a) and as a point spread function depending in (b). The later plot also shows the phase lag of the particle signal, which can be seen in the peaks being shifted to zero-field strength. The magnetic moment is additionally shown in the hysteresis plot as a function depending on the magnetic field (c) and in the Fourier domain as frequency spectra of the odd harmonics (d)

Tracer research needs to deal with two conflicting goals regarding imaging performance. It is known that if the saturation field strength is the lower, the larger the particle cores are. It implies that the spatial resolution in MPI improves with increasing particle volume. However, its inertia and anisotropy depend on the particle size, especially on the non-magnetic shell, resulting in the particle relaxation effects and turn decreasing signal and resolution if the size gets too large.

Finding the optimal particle size and shape is thus one of the core questions in tracer research done both experimentally and in simulation studies. The influence of the diameter of several monodisperse particles on the MPS signal intensity was compared with determine the optimal size for MPI (Ferguson et al., 2013). Tracer with a median diameter of 20 nm and hydrodynamic diameter equal to 30 nm showed the best performance.

Size separation due to forced magnetic field (Arsalani et al., 2021) and centrifugation (Dadfar et al., 2020) is an additional way to increase the MPI signal intensity. In simulation studies, magnetic nanoparticle chains with a specific length improve signal intensity (Zhao & Rinaldi, 2020). In addition to spherical particles and particle chains, maghemite nanoflowers (Fe2O3) (Bender et al., 2018; Karpavičius et al., 2021) provided an excellent MPI performance with a fourfold higher signal than Resovist (Szwargulski et al., 2020). Other authors reported even higher amplification factors (Ziemian et al., 2018), where the signal was a factor of 6.6 higher than Resovist, resulting in improved spatial resolution in imaging experiments. Regarding the particle shape, recently, cubic iron oxide nanoparticles with more than a threefold MPI signal increase compared with VivoTrax were presented (Wang et al., 2020).

Synthesizing multicore nanoparticles can result in a fourfold signal increase compared with Resovist in MPS measurements (Kratz et al., 2018). In vivo angiography of inferior vena cava and aorta of rats with these particles shows the potential of MPI for quantitative assessment of the vascular anatomy (Mohtashamdolatshahi et al., 2020). In another effort, carbon-coated FeCo nanoparticles were presented by Song et al. (2020), revealing a sixfold higher MPI signal than VivoTrax, which has a similar performance as Resovist. In summary, recent progress in particle synthesis has shown that different approaches can increase the particle signal by a factor of 3–6 compared with the gold standard Resovist. It is an important future research topic whether combining the different approaches achieves even higher signals.

Moreover, modification of the particle surface to improve imaging quality has made rapid progress within the last decade. A large class of MPI tracers, including the clinically used tracer Resovist (Reimer & Balzer, 2003), have a coating based on a glucose molecule like dextran or carboxydextran. After intravenous injection, the tracer remains in the blood pool for a short time and is then continuously taken up by the Kupffer cells in the liver and spleen (Guzy et al., 2020; Hamm et al., 1994). While this is helpful for liver applications, it shortens the time window for other applications such as perfusion studies. Typical half-life times of these tracers are about 6 min in human blood (Hamm et al., 1994) and about 5 days in the liver of rats (Lawaczeck et al., 1997). Because of this, there is great interest in increasing the half-life time of MPI tracers and ideally developing tracers that remain in the blood pool for a longer time, thus allowing image vessels and organ perfusion over a long period.

Two different approaches have successfully reached this goal: Initial experiments (Khandhar et al., 2015) attempted to increase the blood half-life time by tuning the surface coating with polyethylene glycol (PEG). MPS experiments in mice revealed a blood half-life time of 19 min. These PEG-coated particles were further improved (Khandhar et al., 2017), with a blood half-life of about 105 min in mice studies. The biodistribution of these particles was tracked in short-term experiments and long-term clearance (Keselman et al., 2017), showing a blood half-life time of 4.2 h, 6.5 days in the liver, and 18.2 days in the spleen of rats. With the 3.4-fold better MPI signal of LS-008 compared with Resovist MPI, in vivo angiography was performed in mice studies showing the potential of real-time perfusion imaging (Kaul et al., 2017). Furthermore, due to the increased MPI signal intensity of LS-008, the spatial resolution improved to 330 μm (Vogel, Rückert, et al., 2019), and the extended blood half-life time allows for perfusion imaging (Kaul et al., 2017). Synomag-D, another PEG-coated tracer with a blood half-life time of 60 min, was used to monitor in vivo intracranial cerebral hemorrhage in mice (Szwargulski et al., 2020).

An alternative approach to adjusting the particle surface for increasing the blood half-life time is to load red blood cells (RBCs) with particles. Practically, the blood retention time was increased with Resovist-loaded RBCs to about 14 days compared with bulk Resovist with a typical retention time of about 1 h in mouse bloodstream (Antonelli et al., 2016). Further MPS experiments with Perimag and Synomag in human RBCs showed a promising potential of Perimag-COOH-loaded RBCs for MPI (Antonelli et al., 2020).

In addition to increasing the half-life time of magnetic nanoparticles, there has also been intensive research in the direction of functionalizing tracers so that specific cells or regions are targeted. These functionalized nanoparticles showed a promising potential in multimodal imaging, both in MPI and MRI (Erathodiyil & Ying, 2011; Hola et al., 2015; Liu et al., 2010).

The tracking and monitoring of ferumoxytol-labeled mesenchymal stem cells (MSCs) was performed with MPI, imaging the total MPI signal decline in vivo within 12 days (Sehl et al., 2020). Later a method for ex vivo counting of cells was presented (Sehl et al., 2020). Bone MSCs labeled with cubic iron oxide nanoparticles were used for stem cell tracking, monitoring the amount and location of the cells over 9 days (Wang et al., 2020). MPI showed to be capable of performing target drug delivery using a peptide CREKA in a mouse breast cancer model (Du et al., 2019). By loading chylomicrons with iron oxide nanoparticles, the in vivo lipoprotein uptake in brown adipose tissue of mice can be quantified using MPI (Hildebrand et al., 2020). In the field of therapeutic imaging, MPI can be used for monitoring the application of hyperthermia. Recently, the combination of targeted drug delivery with nanoparticles and the application of hyperthermia while monitoring the particle location with MPI could be shown (Chandrasekharan et al., 2020; Dadfar et al., 2020; Du et al., 2019; Song et al., 2020).

The toxicity of the tracer materials is a topic that needs addressing. Challenges for clinical translation arise from missing regulatory-approved MPI- specific tracers. Existing SPIOs, developed for other applications, were repurposed for off-label MPI use. Tracers such as ferucarbotran showed minor side effects similar (Onishi et al., 2009).

Compared with gadolinium, MPI tracers have the advantage of being biodegradable, whereas concerns have emerged due to cerebral gadolinium deposits in patients after repetitive applications (Kanda et al., 2014). In contrast to gadolinium, most preclinical trials found a steady decrease in the tracer signal and complete removal or degradation of the SPIOs from the brain parenchyma (Ludewig et al., 2017; Szwargulski et al., 2020). These studies suggest that SPIOs get phagocytosed and digested by macrophages or microglia, followed by incorporating the degraded iron into hemoglobin, which others observed (Bulte, 2019). In the trials mentioned above, no mortality after tracer injections was observed in stroke and bleeding, and the tracer did not increase stroke or bleeding sizes.

In porcine brains, no iron deposition in the brain was observed 9–12 months after intravenous infusion of ferumoxytol (Theruvath et al., 2020). A case–control study found no significant signal differences in brain MRI of children and young adults after exposure to ferumoxytol compared with unexposed children (Iv et al., 2020). Nevertheless, it is necessary to investigate whether SPIOs get entirely removed from the brain parenchyma or increase neurotoxicity when the blood–brain barrier collapses.

One primary clinical concern in imaging techniques regards the risk of immediate hypersensitive reactions after intravenous infusion of contrast agents. Iodine-based contrast agents used for CT scans are long known for risk of acute reactions with approximately 4%–12% for the use of ionic contrast agents and 1%–3% for non-ionic contrast mediums (Bush & Swanson, 1991). The risk of severe and potentially life-threatening events was 0.2% and 0.04%, respectively, leading to the predominating use of non-ionic agents (Bush & Swanson, 1991; Cochran, 2005). On the other hand, gadolinium-based contrast agents show a shallow risk of hypersensitive effects with an overall rate of 0.09% of immediate reactions and 0.005% risk of severe effects (Behzadi et al., 2018).

Safety data drawn from the phase I–III trials of ferucarbotran reported one anaphylactic reaction in 1053 doses (0.09%) and 75 possibly, probably, or definitive drug-related adverse effects, with a majority considered as mild symptoms (Reimer & Balzer, 2003).

Ferumoxytol, another SPIO with an FDA approval for the treatment of iron deficiency anemia, was provided with an FDA boxed warning in 2015 due to the occurrence of 79 anaphylactic reactions, 18 of which had been fatal despite instant medical treatment (US Food and Drug Administration, 2015).

However, approximately 1.2 million doses of ferumoxytol have been administered since the FDA approval in 2009 and the warning release in 2015 (Vasanawala et al., 2016). Four randomized and one non-randomized study reported severe anaphylactic reactions ranging from 0.02% (2/8666 patients) to 1.3% (1 /80 patients). Combined, these studies report a 0,046% rate of severe anaphylactic reactions (5/10,575 patients), whereas the majority of all reported adverse effects was considered mild (Adkinson et al., 2018; Hetzel et al., 2014; Macdougall et al., 2014; Schiller et al., 2014; Vadhan-Raj et al., 2014). A multicenter safety analysis based on a non-randomized observational database of off-label use of ferumoxytol in MRI studies reported no severe or fatal adverse effects in 3215 patients (Nguyen et al., 2019).

Taken together, though not as rare as in gadolinium-based agents, the risk of severe anaphylactic reactions in SPIOs is rare and slightly lower than in iodine-based agents. It can be further reduced by administering the agents in clinical settings with trained personnel, emergency equipment, and cardiovascular monitoring during and after the application.

4 IMAGING OF NEUROLOGICAL DISORDERS 4.1 Potential use of MPI in acute ischemic stroke management

Up to now, multiparametric MRI and CT perfusion (CTP) imaging are the primary modalities for determining the infarct core and penumbra and are commonly used for the image-based acute ischemic stroke treatment decision. Perfusion-weighted MRI (PWI) and CTP are contrast-based imaging techniques (gadolinium and iodine, respectively), which acquire a temporal series of 3D images with a temporal resolution of 1.0–2.5 s of the brain after administration of the exogenous contrast agent. The resulting images display the contrast agent's spatio-temporal (4D) dynamics with reduced and delayed perfusion of the ischemic brain tissue. Due to the high dimensionality and complexity of the 4D image sequences, the PWI or CTP datasets are typically not assessed directly but used to calculate perfusion parameter maps, which are necessary to define the infarct core and the penumbra of the acute ischemic stroke using a wide range of suggested thresholds (Forkert et al., 2013). More precisely, the 4D datasets are used to compute voxel-wise perfusion parameters in the brain, including cerebral blood flow (CBF), cerebral blood volume (CBV), mean transit time (MTT), and the time to the maximum of contrast accumulation (Tmax).

Two main techniques to acquire perfusion-weighted MRI datasets are dynamic susceptibility contrast (DSC) MRI and dynamic contrast-enhanced (DCE) MRI. Briefly described, DCE sequences take advantage of the T1 contrast, while DSC techniques utilize susceptibility effects arising from the application of paramagnetic contrast agents. DCE MRI is sensitive to extravasation effects enabling calculation of the vascular permeability as an additional parameter based on the concentration-time curves. However, since vascular permeability is not used clinically for acute ischemic stroke treatment decision making and T2* effects are considerably stronger than T1 effects, DSC perfusion imaging is the most frequently employed technique for imaging acute stroke patients. Briefly described, DSC measures the susceptibility effects resulting in a T2* shortening of the proton spin relaxation times as the contrast agent bolus travels through the brain.

CT perfusion sequences measure the spatiotemporal changes in density caused by the contrast agent while flowing through the brain. A benefit of CTP compared with DSC is that the contrast concentration is linearly related to the attenuation measured in CT so that no complex correction formulas are needed to transform the measured signal curves to concentration–time curves. The relative or absolute concentration-time curves measured by CTP and DSC-PWI cannot be used directly to determine the perfusion parameters of interest. The measured concentration–time curves are affected by the patient-specific cardiac output function and the contrast agent injection protocol. To correct for these effects, each tissue concentration–time curve is typically deconvolved with the arterial input function (AIF), which is usually identified manually or automatically at the contralateral internal carotid artery or middle cerebral artery before calculating perfusion parameter maps (Winder et al., 2020). While several techniques have been proposed to solve the ill-posed deconvolution problem, the block-circulant singular value decomposition is probably the most often used approach for this purpose (Wu et al., 2003). After deconvolution, the perfusion parameters described above can be determined from the resulting residue curve of each voxel. For example, CBF can be determined by the slope of this curve, MTT by the full-width-half-maximum, CBV by the area under this curve, and Tmax by identifying the time point with the maximum value in the residual curve (Forkert et al., 2014). Slightly different definitions of these parameters have been proposed. After this, the hypoperfused tissue is typically segmented using a global threshold greater than 6 s, while the ischemic core can, for example, be determined using a relative CBF threshold of less than 30% compared with the contralateral hemisphere (Campbell et al., 2011). The volumetric difference between the ischemic core and the hypoperfused tissue is typically assumed to be the tissue-at-risk.

DSC-PWI and CTP both have their benefits and limitations when comparing each other. First, it needs to be highlighted that both techniques use exogenous contrast agents that have been suggested to be potentially harmful to patients in recent studies, for example, showing gadolinium deposition in the brain and other organs with unknown long-term effects (Guo et al., 2018; see chapter 3). DSC-PWI has a much better signal-to-noise ratio (SNR) than CTP, which usually leads to smoother-looking perfusion parameter maps and better segmentation of the hypoperfused tissue. In contrast, considerably more spatial and temporal smoothing is needed to achieve similar results using CTP datasets. However, excessive smoothing can also bias the absolute perfusion parameter calculation. On the other hand, CT scanners are much more widely available in the emergency room setting, allowing faster image acquisition, which is essential given the time constraints of acute ischemic stroke diagnosis (“time is brain”). Furthermore, CT scanners are also significantly cheaper than MRI scanners, so that they are more widely available in small and rural hospitals. While the spatial resolution of CTP has increased considerably over the last decade and full brain coverage is possible with the newest generation of CT scanners, the usage of ionizing radiation needs to be considered as a significant health concern, especially in younger patients, given that a typical CTP series consists of 30–60 full 3D CT head scans. Furthermore, iodine-based contrast agents bear the risk of severe adverse allergic reactions (Bush & Swanson, 1991; see chapter 3). Both imaging sequences have additional drawbacks due to modality-specific artifacts during image acquisition. In CTP, metal artifacts can render the scans unusable for perfusion map quantification. DSC-PWI sequences can be affected by various artifacts such as contrast agent saturation, distortion, and eddy current effects, just to name a few. From a computational perspective, image sequences with a higher temporal resolution, and better signal-to-noise ratio than what can be achieved with DSC-PWI and especially CTP would be highly favorable. This would allow an improved calculation of the perfusion parameter maps with improved spatial and temporal accuracy. The determination of infarct core and the salvageable penumbra has gained additional importance in recent years. The accuracy and reliability of CT and MRI perfusion have recently been critically discussed (Huang et al., 2017). Defining the viable tissue through perfusion imaging is necessary to select optimal candidates for reperfusion therapy. However, the low temporal resolution of greater than 1 s per 3D image in standard CT and MRI perfusion imaging can significantly influence the quantification of salvageable tissue depending on the method and threshold parameters used, which might ultimately exclude patients of an optimal stroke treatment (Forkert et al., 2013). Multiband EPI acquisition provides a temporal resolution of less than 1 s, but multiband MRI scanners are expensive, and reconstruction of the parameter maps is time-consuming.

Due to the high temporal resolution coupled