Combining magnetic particle imaging and magnetic fluid hyperthermia for localized and image-guided treatment

Abstract

Magnetic fluid hyperthermia (MFH) has been widely investigated as a treatment tool for cancer and other diseases. However, focusing traditional MFH to a tumor deep in the body is not feasible because the in vivo wavelength of 300 kHz very low frequency (VLF) excitation fields is longer than 100 m. Recently we demonstrated that millimeter-precision localized heating can be achieved by combining magnetic particle imaging (MPI) with MFH. In principle, real-time MPI imaging can also guide the location and dosing of MFH treatments. Hence, the combination of MPI imaging plus real time localized MPI–MFH could soon permit closed-loop high-resolution hyperthermia treatment. In this review, we will discuss the fundamentals of localized MFH (e.g. physics and biosafety limitations), hardware implementation, MPI real-time guidance, and new research directions on MPI–MFH. We will also discuss how the scale up to human-sized MPI–MFH scanners could proceed.

Keywords:

• Magnetic particle imaging
• magnetic fluid hyperthermia
• localized heating
• image-guided treatment
• magnetic nanoparticles

1. Introduction

1.1. cancer treatment methods and tradeoffs

The preferred treatment of a primary cancer is surgical resection, whenever feasible. However, for some tumors, including metastatic tumors, physicians often recommend radiation therapy, chemotherapy, and, increasingly, immunotherapy. Of course, all cancer treatments must balance benefits against risks [1]. For example, radiation therapy and chemotherapy have been shown to reduce the growth rate of tumors but these can also damage healthy tissues that are close (radiotherapy) or far (chemotherapy) from the tumor. Surgery has risks of infection and bleeding and is not always curative for metastatic tumors, some of which are poorly visualized on imaging studies like positron emission tomography (PET), computed tomography (CT), and magnetic resonance imaging (MRI).

To minimize morbidity and mortality of more invasive treatments, hyperthermia has evolved to be an active research area. Here, heat is used to irreversibly damage the pathological targets. Hyperthermia has been used to treat heart arrhythmia, cauterize the endometrial wall, cauterize arteriovenous malformations to prevent hemorrhagic strokes, and to treat metastatic and recurrent tumors. Heat can be generated in many different ways, including use of ultrasound, radiofrequency (RF) magnetic fields [2], and lasers. However, many of these methods have significant depth or focus limitations. For example, laser light is typically used for surface treatments due to penetration limitations. Small RF catheters can apply localized heating, but with additional invasiveness. The use of magnetic nanoparticles (MNPs) coupled with alternating magnetic fields, known as magnetic fluid hyperthermia (MFH), has shown great promise in cancer research with the potential to overcome many of the limitations presented by other hyperthermia techniques.

1.2. Outline of the review

Below, we first review MFH technology as a cancer treatment (Section 2), followed by MPI technology (Section 3), and fundamentals of MPI–MFH (Section 4). The efforts of implementing combined MPI and MFH systems, including gradient and RF coils, are then discussed (Section 5). We also review particles tailored for both MPI and MFH (Section 6), as well as the limitations on specific absorption rate (SAR) and peripheral nerve stimulation (PNS) of MPI and MFH (Section 7). In the end, we discuss future perspectives, including the possibility of monitoring cell viability during MFH using MPI and the effort of scaling up to human-size MPI-MFH systems (Section 8).

2. MFH as a cancer treatment method

Magnetic fluid hyperthermia is a minimally invasive form of thermal therapy. Here, we transfer heat to the body by inducing RF energy primarily within pre-injected MNPs [1,3,4]. Energy deposition occurs when MNPs cycle through their magnetization (M–H) curve due to the alternating current (AC) magnetic drive field. The energy deposited is directly related to the area traversed under the M–H curve.

2.1. Advantages of MFH

Compared to other cancer treatment techniques, the advantages of MFH are: (1) no ionizing radiation; (2) no penetration depth limitation, which means it can be used to treat anywhere in the body; (3) heating is highly localized to MNPs and care is taken to minimize background tissue heating; (4) it is a noninvasive method, which could offer hope for patients with tumors that are in locations that are simply too dangerous to remove surgically; and (5) it is possible to treat the entire tumor. In addition, some of the biomedical advantages of MFH are discussed below.

Hyperthermia is of interest in combination with chemotherapy and radiotherapy. In magnetic hyperthermia, MNPs can be engineered for drug encapsulation and the application of MFH can magnetically trigger controlled drug release [5–7]. Moreover, radiotracer binding can allow for hyperthermia and radiotherapy combinations [8]. MFH can be administered as an adjuvant therapy after tumor sensitization to ionizing radiation or chemotherapy [9,10], having also the potential to overcome drug resistance when applied as a sequential treatment strategy [11].

The clinical significance of hyperthermia continues to increase because it can also elicit immunogenic cell death, leading to an anti-tumor immune effect [12]. When treating primary tumors with MFH, the innate and adaptive immune response that is mounted can also have an effect on metastatic lesions [13]. Independent studies have shown that MFH induces an immune response against cancer, in which both treated and untreated contralateral tumors disappear [14] and resistance to secondary tumor rechallenge emerges [15] with MFH performed at 43 °C but not at 41 °C or above 45 °C [16]. It is important to note that ablation temperatures (>45 °C) might cause complete tumor tissue death, which will not induce an anti-tumor immune response. This underscores the importance of treatment and dose control to enable the benefits of MFH.

The MFH technology has been shown to be effective in the clinics. NanoTherm® obtained European Medicines Agency (EMA) approval in 2010 for their use in thermal cancer therapy and have an Investigational Device Exemption (IDE) approved by the Food and Drug Administration (FDA) for thermal treatment of prostate cancer [5,17]. Promising clinical results for the use of NanoTherm® therapy in glioblastoma multiforme have been reported [18]. An overall increase in survival of 13.4 months following diagnosis of first tumor recurrence and 23.2 months after primary tumor diagnosis were reported in comparison with historical controls of 6.2 and 14.6 months, respectively [5,18]. Success in these clinical trials can open the path to new approvals and clinical technology development.

2.2. Challenges of MFH

One of the biggest challenges in clinical application of MFH is off-target accumulation and heating. MNPs accumulate naturally in off-target organs (e.g. liver, kidneys, bladder, and spleen) [19]. This could lead to unintended heating damage in off-target tissues [20]. Current MFH approaches cannot spatially localize heating to the tumor while leaving MNPs in the liver or spleen at body temperature. This is because focusing of the excitation field (100 − 1000 kHz) is fundamentally limited by diffraction. In fact, regardless of the number of external coils employed, one cannot focus VLF energy to a spot size smaller than roughly half the in vivo wavelength (17 − 150 m), which is unfortunately not helpful for clinical applications [21,22].

Another challenge with MFH is real-time guidance. Physicians would prefer to treat under real-time guidance, with a high resolution real-time imaging modality that can identify the location of the MNPs and the MFH heating volume. Unfortunately, physicians currently have no choice but to use CT and MRI to guide tumor treatment with MFH [18,23]. These imaging methods are typically not real-time, so treatment plans often must rely on images taken prior to the treatment. However, the MNP distribution might change during the process.

2.3. Ideal MFH and MPI solutions

An ideal MFH platform would have the following improvements: (1) user-defined arbitrary and high-resolution localization of hyperthermia, allowing for pinpoint treatment of the entire tumor without damaging healthy tissue; and (2) real-time imaging and sensing to guide MFH heating. This review shows the promise of MPI hardware to perform both of these tasks soon. We recently demonstrated success in using MPI hardware to simultaneously localize and guide MFH within the body. The hardware required is similar to that used in standard magnetic particle imaging (MPI) scanners. Currently, animal MPI–MFH systems are available. Clinical scale MPI has been reported for head/brain applications, and a clinical magnetic field applicator (NanoActivator®) is available and in clinical use.

3. Magnetic particle imaging

3.1. Advantages of MPI among other imaging modalities

Magnetic particle imaging is a novel tracer imaging modality introduced by Gleich and Weizenecker [24]. This imaging modality directly detects the flipping of the intense magnetization of MNPs, specifically superparamagnetic iron oxide nanoparticles (SPIOs). MPI has been proven a robust, rapid and quantitative imaging method [25–30]. One of its most promising advantages is that MPI can detect the MNPs anywhere in the body with no depth attenuation, and with zero clutter from background tissue. This means that MPI is a high signal to noise ratio (SNR) imaging platform like nuclear medicine [24,31]. MPI has demonstrated promise for applications such as angiography [32,33], stem cell tracking [34–36], brain perfusion [37,38], lung perfusion [39], lung ventilation [40,41], cancer imaging [42], gut bleed detection [43], remote magnetic actuation and magnetic catheter steering and guidance [44–47], and localized MFH [21,22,48,49]. Selected MPI images are shown in Figure 1.

Figure 1. MPI can provide radiation-free, highly sensitive, and high resolution images. (a) Sensitive detection of gastro–intestinal bleed (Reproduced with permission from Yu et al. [43]. Copyright (2017) American Chemical Society.) (b) Evaluation of inhaled drug therapeutics–MPI is not affected by air-tissue susceptibility interfaces and can even image in lungs (Reproduced with permission from Tay et al. Theranostics 2018 [40]). (c) Hemorrhage diagnosis in TBI animal model (Reproduced with permission from Orandoff et al. 2017 Institute of Physics and Engineering in Medicine, IOP Publishing [37]).

Fundamental advantages of MPI include [50,51]:

• MPI has ideal penetration and SNR: MPI uses low frequency magnetic fields with zero depth attenuation and zero clutter from background living tissue.

• MPI has no view limitations: MRI is unreliable in the lungs due to short T2* near ± 5 ppm air-tissue field variations [52]. MPI tolerates 5% field variations, whereas MRI typically requires 1–5 ppm homogeneity. Hence, MPI works as normal in bones, in lungs—anywhere. There is zero de-phasing of the MPI signal in the lungs, unlike T2* effects in MRI. Also, ultrasound waves cannot penetrate through lung tissue or bones.

• MPI signal is sensitive, linear, and quantitative: The MPI induction signal is an accurate count of the number of MNPs in a voxel [34,35,53]. The MPI signal never saturates, even at maximum clinical concentration of 2 mM iron [54]. MPI shows 1 micromolar sensitivity (about 200 cells with 27 pg/cell labeling efficiency [34,35], more sensitive than all “positive contrast” imaging methods [55,56]).

• Tracers are safe and persistent: The MNPs used in MPI are SPIOs which are regarded as biocompatible and have shown to biodegrade in vivo. Unlike RBC-Tc99m, which is limited by its 6-h radioactive half-life, SPIOs blood circulation half-life will depend on the particle properties and their surface coating [57]. However, SPIO tissue half-life is basically infinite when accumulated in target tissue or when it is directly injected into the tumor. Zheng et al. [35] demonstrated that SPIOs can be tracked for more than 80 days with minimal MPI signal loss.

3.2. MPI fundamentals

Magnetic particle imaging works by detecting the nonlinear magnetization of the flipping MNP tracers. Strong magnetic field gradients magnetically saturate the MNPs everywhere outside of the field-free region (FFR). Therefore, only those MNPs in the FFR can respond to a small external excitation field. MPI uses an inductive pickup coil to detect the flipping of magnetization of these MNPs. Due to the nonlinear response, MNPs have rich harmonics other than the fundamental frequency, which makes it possible to differentiate MNPs signals and induced direct feedthrough interference. To collect the image data, MPI scans a 3D volume by rastering the FFR around (see Figure 2). Image reconstruction can be performed using the X-space approach by gridding the MPI signal to the instantaneous location of the FFR [31] or by a frequency domain approach by solving the inverse problem with a calibrated system function [26]. All living tissues (including ferritin and hemoglobin) are magnetically weakly linear (susceptibility, $\left|\chi \right|<$ 10 ppm) [52]. Since MPI scanners only pick up harmonics of the 20 kHz drive frequency, linear magnetic materials (like water, fat, muscle and bone) produce no harmonics. Background tissues are linear magnetic materials and so they emit only at 20 kHz—which is rejected before amplification and digitization to avoid “direct feedthrough”. Hence, all human (and murine) tissue is completely invisible to MPI and MPI is thus similar to PET/SPECT, which only detect radiotracers. Similar to PET-CT and PET-MRI scanner designs, MPI scanners can also be integrated with CT [58] and MRI [59].

Figure 2. MPI fundamentals and systems. (a) A strong magnetic field gradient forms a sensitive point known as a Field Free Point (FFP). The FFP can be moved in a scanning trajectory to cover the imaging field of view. (b) SPIOs’ magnetization demonstrates nonlinear behavior as a function of the applied field, which can be modeled using a Langevin function ((a) and (b) are reproduced with permission from Patrick et al, Advanced Materials 2012, John Wiley and Sons [81]). (c) and (d) are the hardware setups of MPI and MPI–MFH (Reproduced with permission from Tay et al., Copyright 2018 American Chemical Society [22]).

4. MPI–MFH

4.1. Localized MFH with gradient-based localization techniques from MPI

4.1.1. Fundamentals of gradient-based localization

A long-standing challenge in the MFH field is localizing heating to a precise target deep within the body [60]. This is important given the nonspecific uptake of MNPs to the liver and spleen, which typically exceeds the specific uptake to the tumor [60]. Unfortunately, it is not possible to focus magnetic fields with frequencies on the order of 300 kHz to avoid heating the liver because the wavelength λ of such fields is greater than 100 m in vivo [60]. The fundamental spatial resolution of any focusing system is limited to roughly λ/2. Small surface coils can localize the magnetic field using the so-called “near-field” or quasi-static field fall off. However, this fall off field does not allow for pinpoint localization deep inside a human or animal. Given these fundamental restrictions on localization, current MFH methods pose risks of (1) heating off-target tissue/organs (e.g. liver) when attempting to heat only a tumor, and (2) non-uniformly heating the tumor, thereby limiting the ability of MFH to treat the entire tumor.

To address this important technical challenge, several research groups have demonstrated experimentally that the MPI gradient field can localize MFH heating [21,22,61–64]. Despite using a 300 kHz magnetic field, we and others have demonstrated localized heating to millimeter-scale resolution. To achieve this, we exploit the nonlinear response of MNPs to two superimposed magnetic fields: a strong, static gradient field and a 20 kHz uniform field. The nonlinear heating of MNPs allows for pinpoint heating only near the null field point of the static gradient field. If we place the null field point of the gradient over a tumor just a few millimeters away from the liver or spleen, the combination allows for heating MNPs in the tumor without heating them in the liver or spleen, as shown in Figure 3.

Figure 3. Localized selective heating using MPI–MFH. MPI gradients enable selective heating. The figure shows that the temperature increased only at the selected area. The animal liver and spleen were not affected when the gradient field was turned on. (Reproduced with permission from Tay et al., Copyright 2018 American Chemical Society [22]).

The dominant heat generation process in MFH is that of relaxation in the MNPs under AC magnetic field cycling [1]. Magnetic field cycling is also used for signal generation in MPI, where the induced electric signal from the changing magnetization of particles is measured [24,31]. The physics that underlies MPI and MFH are therefore very similar in that they both exploit the nonlinear magnetization response to the oscillating magnetic fields. This makes it possible to transfer MPI gradient localization techniques to MFH.

Particles in the FFR can be freely magnetized and their magnetization manipulated by an external magnetic drive field. Particles outside the FFR, however, have their magnetization saturated by the gradient field. Hence, the magnetization of MNPs outside the FFR moves back and forth, but never departs from saturation. Hence, these particles outside the FFR generate almost zero induction in the external pickup coil. Indeed, the pickup coil only “sees” signal from the FFR. The behavior of MNPs can therefore be fully controlled by the combination of the applied gradient field and the AC drive field. Only the particles within the FFR can generate an MPI signature, and only particles in the FFR can generate heat via magnetization-induced losses. The spatial resolution of heating scales proportionately to the MPI spatial resolution since the localization mechanism is similar. In this manner, an MPI–MFH system can provide targeted heating with high resolution at arbitrary depths in the body.

4.1.2. Spatial resolution of MPI and MPI–MFH

The adiabatic point spread function (PSF) of x-space MPI was investigated by Goodwill et al., who showed that the full-width half-maximum (FWHM) of the PSF is inversely proportional to the gradient strength (G) and particle volume (d³) [31]: (1) $$\Delta x\approx \frac{24k_{B}T}{{\mu }_0\pi M_{\mathit{sat}}}\frac{1}{G{d}^3}$$(1) where k_B is the Boltzmann constant, T the temperature, μ₀ the permeability of free space, d the particle diameter, and M_sat the saturation magnetization of the magnetic particles. Importantly, this resolution for the x-space reconstruction is identical to the spatial resolution predicted by the System Matrix reconstruction method in MPI [26,65]. This equation holds true for small particles where relaxation effects can largely be ignored. For larger particles (>25 nm core diameter), relaxation effects become prominent and have to be considered [66]. Starting from the Debye relaxation equation [67], Croft et al. [68,69] derived the non-adiabatic PSF in x-space. This PSF can be modeled as a convolution between the adiabatic PSF and a relaxation kernel r(t), (2) $$r(t)=\frac{1}{\tau }\text{exp}\hspace{0.17em}\left(-\frac{t}{\tau }\right)u(t)$$(2) where τ is the field-dependent relaxation time constant and u(t) is the Heaviside step function. The relaxation-dependent PSF of MPI as shown in Figure 4(a) was also experimentally observed by Tay et al. using an arbitrary waveform relaxometer [66]. The highest spatial resolution of approx. 1.7 mm was observed for a 25 nm core size MNP under a 7 T/m gradient field.

Figure 4. Spatial resolution of MPI and MPI–MFH: (a)Measured MPI PSF using an AWR (Reproduced with permission from Tay et al., 2017 Biomedical Physics & Engineering Express, IOP Publishing [66]); (b) Simulation showing the SAR resolution of MPI–MFH under different frequencies of the oscillating magnetic field (Reproduced with permission from Dhavalikar et al., Journal of Magnetism and Magnetic Materials 2016, Elsevier [70]); and (c) the measured heating resolution of MPI–MFH using an optical probe (Reproduced with permission from Tay et al., Copyright 2018 American Chemistry Society [22]).

The spatial resolution of particle SAR for MPI–MFH was simulated by Dhavalikar et al. [70]. The MNPs dissipate heat through relaxation losses which can occur due to Neel or Brownian relaxation. The simulation modeled the SAR of the MNPs in a gradient field under different driving waveform conditions (Figure 4(b)). The model considered a 1 D system with a linear gradient field. By using the phenomenological magnetization equation derived by Martsenyuk, Raikher, and Shliomis [71–73], the model takes field-dependent relaxation into consideration. SAR correlates directly with the particle energy dissipation. To get the heating resolution, one has to take into account the thermal diffusion effect. The magnetic resolution $\Delta H$ (mT) in Figure 4(b) can be translated to x-space using (3) $$\Delta x\propto \frac{\Delta H}{G}$$(3)

The heating resolution therefore improves with increasing gradient strength (G). As shown in Figure 4(b), $\Delta H\approx$ 24 mT for excitation field at 300 kHz, which translates to a $\Delta x\approx$ 3.4 mm in x-space under a 7 T/m gradient field. Figure 4(c) showed the measured heating PSF of MPI–MFH under a 354 kHz excitation field. Both SAR and the effect of thermal diffusion in the sample were covered in the measured experimental data. The temperature was measured using an optical probe [22]. A 7-mm spatial resolution was observed using a 2.35 T/m field free line (FFL) scanner. Tay et al. developed a predictive algorithm to calculate the temperature map by convolving the localized SAR image (derived from the MPI image) with the measured heating resolution in temperature [22]. Because MPI can provide real-time information about the MNP distribution, the effects of particle diffusion during MFH on temperature monitoring could be mitigated. The reader can find further discussion on MPI induced temperature sensing in Section 4.2.

4.2. Real-time guidance of MFH by MPI

Another advantage of combining MPI and MFH is that MPI can provide real time feedback for safer and more refined therapy. This is not possible using current MFH techniques, which rely on CT or MRI as imaging tools [18,23,60]. Nowadays, it is not uncommon for physicians to rely on days-old images to inform their treatment plans. Murase et al. recently showed that MPI can be used to predict the effect of subsequent MFH treatment, owing to the common physics that underlies the generation of both the MPI signal and MFH heating [48]. MPI could provide continuous monitoring of tumor position, on-the-fly quantitation of SAR deposition or temperature [74,75], and real time assessment of treatment success. Both the MPI signal and MFH heating efficiency are relaxation dependent, a property that cannot be pre-imaged or calibrated for using CT or MRI. For instance, relaxation can change due to the temperature rising during hyperthermia and binding events when using targeted tracers. Such nuances in MFH treatment call for a real time monitoring approach, thus precluding current MFH techniques that are open-loop and require invasive temperature probes [76].

Image guided MFH will help determine the in vivo location of MNPs in the tissue of interest and apply AMF to achieve localized heating without damaging nearby healthy tissue [21]. This is of great potential clinical impact because whole body MFH using coils surrounding the patient will heat off-targeted tissues that will cause unwanted side effects, while regional MFH is limited to surface tumors. However, using MPI guided MFH one can image the biodistribution of MNPs and confirm their anatomical location [21] with previously acquired MRI or CT images. As such, real time MPI will help to accurately select the treatment area with millimeters of resolution while sparing off-target tissues, and perform MFH only to the cancer tumor (Figure 5).

Figure 5. Comparison of MPI–MFH among other magnetic hyperthermia techniques: (a) Whole body MFH will heat all locations where MNPs are accumulated including healthy tissue; (b) Surface coils can be used to target surface tumors but not sites deep in the body; (c) MPI–MFH can target any site in the body using the FFR, including those deep in the body with high resolution [21]. (Reproduced with permission from Hensley et al., 2017 Physics in Medicine & Biology, IOP Publishing [21]).

The advantage of quantitative imaging of irons or MNPs in vivo at any place in the body makes MPI a great imaging modality for MFH dosage planning. Because the thermal energy deposited is linearly related to the amount of MNPs, by accurately quantifying the amount of MNPs in vivo, the expected thermal energy deposited can be directly calculated with the precalibrated SAR of the MNPs [22]. Tay et al. showed SAR deposited and the temperature increasement are linearly proportional to the MPI image intensity experimentally [22]. This result proved that MPI can be very useful in clinical situations for dosage planning. Tay et al. also developed a predictive algorithm for the in vivo distribution of heating. The localization heating PSF is precalibrated experimentally (Figure 4 (c)). The temperature image can be derived by convolving the localized SAR image (derived from the MPI image) with the measured heating PSF in temperature. More detailed information can be found here [22].

Though more dosage and temperature can be accessed by calculation, in vivo temperature feedback can provide the most accurate assessment of thermal therapy and dosage [22]. Researchers are currently investigating MPI as noninvasive real time thermometry for MFH treatment, specifically by detecting the temperature-dependent changes of MNP response under the alternating magnetic field. The measured temperature represents that of the microenvironment surrounding the MNPs. Currently, there are two predominantly investigated methods, one of which relies on the relationship between relaxation time and temperature [77]. Probing relaxation information has been investigated by some researchers [68,69,78]. Relaxation can also be directly measured using the Pulsed MPI method [79]. The other commonly studied temperature sensing method is based on the temperature-dependent magnetization of MNPs described by the Langevin function, which can be detected through the changes in the magnetization response [74]. Both methods show great temperature accuracy in vitro, specifically, 0.42 °C from the relaxation-based method and 0.3 °C from the Langevin function based method [74,77]. Future in vivo studies depicting the resolution of the temperature map will be crucial in developing novel clinical applications for MFH.

4.3. Thermal dose delivery strategy of MPI–MFH

Current MFH excites all the particles at the same time and it is very challenging to realize equal treatment (or maintain equal thermal dose) for the whole treatment area because of the inhomogeneity of the distribution of the MNPs administered. This will lead to insufficient thermal dose for some treatment areas with lower particle concentration and overdose (which is harmful to the normal tissue) for some areas with high particle concentration. In some cases, this might lead to thermotolerance issues [80]. However, MPI induced localized heating technique can selectively excite MNPs with $\approx$ 2 mm linear spatial resolution (with 7 T/m gradient field). In other words, MPI–MFH can deliver the thermal dose equally to the whole treatment area based on the particle concentration in a specific voxel (which can be obtained in vivo, in real time using MPI). This would also be a very important thermal dose delivery technique when the treatment areas are surrounded by different bioenvironments (e.g. the areas close to the main blood vessels). Also, the localized heating can also help get rid of heating from the particle accumulated organs (e.g. liver). The voxel-specific thermal dose delivery can be realized by tuning the amplitude of the excitation pulse and/or the duration of the FFR residing in a specific voxel.

As we discussed in previous sections, MPI could provide a real time image of MNP distribution and the signal intensity of the image is directly proportional to the amount of the MNPs. With a precalibrated relationship between the amount of nanoparticles and temperature, MPI offers estimated temperature information. That said, MPI image could provide real time guidance for the thermal dose delivery which is currently challenging using other image-guided techniques.

Current CEM43 method estimates the thermal dose for effective treatment. However, the heat sensitivity differs from tumor to tumor, tissue to tissue, and cell to cell. MPI could probe the real time viscosity changes of the cell. Given the hypothesis that there is a viscosity change in the cell during an apoptosis process, MPI could be used for real time cell viability sensing. We discussed this in detail in section 8. With MPI cell viability sensing, MPI–MFH could soon promise MFH community with real time cell viability tracking during and after the treatment. This feedback will help make a more accurate thermal dose for effective treatment.

5. Implementing real-time MPI–MFH

5.1. Implementation of a localization gradient field

The localization technique using a strong magnetic field gradient is well established in MPI [24,31,81–83]. A FFR is generated by two opposing permanent or electromagnets. Depending on the shape of the magnets, the FFR could be a field free point (FFP) or a FFL [82,84]. As shown in Figure 2(c), two magnets were installed outside of the RF excitation coil. A mechanical frame is necessary to constrain the two opposing magnets. An FFL scanner relies on motors to move the sample to get an MPI image [84]. However, extra coils (powered by extra power amplifiers) have to be designed and installed to drive the FFP in a 3 D FFP scanner [82]. The system setup of an FFL scanner is much simpler, with less cost and less power consumption than an FFP scanner. During treatment, the FFL is set to be perpendicular to the frontal plane of the human body and therefore, this could avoid affecting other MNP accumulated organs such as the liver and spleen. Since MFH also uses a RF coil to excite MNPs, implementing a localization magnetic gradient field in a MFH system can directly follow the setup of an MPI scanner. Recently, we demonstrated in vivo localization of MFH using a field gradient as shown in Figure 2(d) [21,22]. An FFL with a gradient of 2.35 T/m was achieved using two rectangular permanent NdFeB magnets, which led to a 7-mm heating resolution in MFH. The authors expect 2 mm heating resolution by employing a 7 T/m gradient field. This will be adequate to achieve refined therapy and avoid heating surrounding healthy tissue/organs while MFH. In clinical practice, a gradient field generated by electromagnets will be more useful, enabling flexible spatial resolutions in MFH for different applications or treatment regions.

5.2. Hardware strategy for combined MPI and MFH

An ideal real-time theranostic platform will allow real-time visualization of the pathology and treatment region, such that imaging and therapy are carried out simultaneously [22]. Because the MNPs also generate MPI signal during heating, real-time simultaneous MPI-therapy should be possible. To achieve this, the MPI imaging and MFH heating drive coils, which are currently separate, should be combined in one scanner with the addition of a receive coil to pick up MPI inductive signal. Ideally, MPI and MFH can work at the same RF frequency, f₀ = 354 kHz. However, this poses impractical constraints for the front-end MPI electronics to obtain adequate harmonics for MPI image reconstruction. MPI currently uses commercial preamplifiers with a 1 MHz bandwidth. That means only 1 harmonic (given f₀ = 354 kHz) is received after the initial low-pass filter, insufficient for image reconstruction. A simple way to realize real-time MPI–MFH is generating two fundamental frequencies $f_{0,1}$ = 20 kHz (for MPI) and $f_{0,2}$ = 354 kHz (for hyperthermia) simultaneously in the RF coil. This two frequency excitation method has already been shown by Goodwill et al. [85] to generate successful MPI images, with possible SNR improvements offered by better noise matching. A band-pass filter has to be designed to filter out both the lower and higher frequency feed through, limiting the detectable MPI harmonics to $n=f_{0,2}/f_{0,1}-1,$ or 16 for this case. This will still allow for comparable image quality to the 20 harmonic MPI images commonly used for commercial magnetic particles.

One practical limitation of this combined imaging-therapy system is that real-time imaging can only occur in the region being treated because the FFR for both imaging and localized heating therapy has to be parked at the treatment area. This is less relevant once the region-of-interest was initially imaged to be used to restrict the tumor vicinity. However, if the larger field-of-view image is required, one can separate the process of imaging and treatment in time as proposed by ZW Tay et al. [22], keeping 5% of time for imaging the larger field-of-view and another 95% of the time for heat treatment. This strategy is fast enough to monitor the SPIO migration in the tumor (usually on the timescale of hours [42]) given the fast scanning time of MPI, where up to the rates of 46 frames per second was reported [86].

6. Tailoring MNPs for MPI and MFH

Magnetic nanoparticles used in MPI and MFH, unlike ferromagnetic materials, have zero net magnetization when the applied external field turns to zero at low frequency. However, at higher frequencies, where relaxation time constant of SPIO is higher than the magnetic field variation period, there is a lag between magnetization and the external oscillating magnetic field [87,88]. This results in an enclosed area in the M–H curve of these MNPs, which results in heat release to their environment [5,87,89,90]. Commercial MNPs (e.g. Feridex and Resovist) have been used in the past as negative contrast agents in MRI, where they produce a local reduction in signal [5,17]. However, these particles did not gain widespread clinical use for this purpose and were eventually withdrawn [5,91], even though they have shown a safe clinical history. Now, these MNPs have been widely used for different MPI and MFH applications. However, these commercial nanoparticles have not been tailored as MPI imaging agents or for MFH applications.

To improve MPI sensitivity and resolution, efforts can also be made in the materials and particle synthesis field. Common synthesis methods such as thermal decomposition will produce MNP with mixed iron oxide phases, crystal defects, and poor magnetic properties, that will ultimately affect the MNP performance. However, Unni et al. [92] showed that by the addition of molecular oxygen as one of the reactive species in the reaction we can obtain single-crystalline MNPs with few defects and similar magnetic and physical diameter distributions. Similarly, for MFH applications, the ideal particles will have a narrow size distribution, phase purity, colloidal stability, and optimal magnetic properties [92,93]. Particles that are tailored for MPI–MFH can be characterized using an arbitrary waveform relaxometer (AWR) (Figure 6) [96]. It is a 1D version of an MPI scanner and can operate at any waveform within a large frequency range (DC-400kHz). Therefore, AWR can get the PSF information from particle, which represents the spatial resolution. Experimental data by Ferguson et al. [97] shows that SPIOs with 25 nm in diameter that were tailored for MPI showed up to 3× greater SNR and better spatial resolution than commercial tracers.

Figure 6. Arbitrary waveform relaxometer (AWR) for characterizing magnetic nanoparticles for MPI and MFH: (a) Prototype of AWR developed at UC Berkeley. (b) The AWR covers all possible driving waveforms (in both frequency and field amplitude) considered safe for human scanning, and can be used to test MNP performance. This enables comprehensive driving waveform optimization. In contrast, conventional VSM [94] and AC susceptometry [95] cannot cover the entire driving waveform range of interest for MPI. (c) The AWR can characterize the nanoparticle PSF without the need for an MPI scanner. (Reproduced with permission from Tay et al., Scientific Reports 2016, Springer Nature [96]).

Initial work developing hardware and theory for MPI assumed that the nonlinear magnetization response of the tracer is instantaneous, meaning that there is no lag between change in applied magnetic field amplitude and nanoparticles magnetization. However, MNPs in time-varying magnetic fields respond with characteristic finite relaxation times and these can be significant for MPI tracers. For x-space MPI, Croft et al. showed that magnetic relaxation will cause a shift in the peak of the PSF that results in blurring and loss of SNR [68]. As such, it is important to consider the relative contributions of the Brownian (physical particle rotation) and Néel (internal dipole rotation) relaxation mechanisms to the response of MNPs in applications such as MPI and MFH [92]. The dominating mechanism depends on a complex function of the size and shape of the particles, their magnetic properties, and the conditions of the applied field [5,92]. Importantly, the characteristic relaxation time generally decreases with increasing amplitude of the applied AMF, although at different rates for Brownian and Néel relaxation [98]. Furthermore, the two relaxation mechanisms have distinct dependence on nanoparticle size. Thus, depending on the nominal size and breadth of the size distribution of the nanoparticles in a tracer, there could be a mixture of Brownian and Néel relaxation mechanisms at work. While there has been both computational and experimental work evaluating the effects of MNP relaxation on MPI, this is still a new field and there are open questions that remain. Further research is needed to elucidate how magnetic fields used in MPI influence MNP behavior, with consideration of nanoparticle size distribution and interactions. An important parallel can be drawn here to the field of MFH, where there is active and ongoing research on how mechanisms of nanoparticle relaxation impact heating rates and cell death. In MPI, it will be important to know how the mechanism of magnetic relaxation influences MPI performance in real biological environments.

7. MPI and MFH safety: SAR and PNS limitations

Electromagnetic safety assurance will be crucial for any human studies or treatments with MFH. Ensuring the electromagnetic fields pose no harm to patients is the responsibility of the FDA in the USA and the International Electrotechnical Commission (IEC) in the European Union. Fortunately, the MFH industry can rely on decades of electromagnetic safety research based on similar magnetic fields employed in human MRI scans. Specifically, the FDA and IEC have established guidelines on preventing excess patient heating (called SAR) and PNS. Heating SAR is caused by conductive joule heating due to ions flowing in human tissue under the action of an induced electric field. Likewise, PNS is also caused by induced E-fields. Here, the PNS occurs when an induced voltage exceeds the activation potential of a nerve. Because both SAR and PNS are safety concerns, the FDA and the IEC prescribe limitations on the induced E-field, which in turn limits the concomitant B-field attributes (e.g. FOV, amplitude, frequency, duty cycle, etc). SAR is limited to 4 W/kg to ensure small temperature increases in the body. And PNS is generally avoided by ensuring that the rate of change of the B-field remains limited to 20 T/s.

The interested reader should consult the many excellent papers written on electromagnetic modeling of both SAR and PNS on MRI. For PNS in MRI, readers should refer to Irnich and Schmitt [99] and Chronik and Rutt (2001) [100] in which magnetostimulation thresholds are discussed. Also, for SAR in MRI, readers should consult guidance issued by the US FDA [101] and the IEC [102]. For SAR and PNS in MPI readers should consult Saritas et al. [103] and Graeser et al. [104], in which MPI magnetostimulation limits and feasibility of human scale MPI are discussed, respectively. In addition, for SAR in MFH readers can refer to Kozissnik et al. [105] in which they underscore the need for new studies to determine the effect of AMF in normal tissue under MFH conditions to adequately determine MFH safety limits.

Manufacturers have great hope for the future of MFH therapies in humans. Indeed, one commercial vendor (MagForce, Berlin Germany) has already received human approval. Specifically, their website states that “Since 2011, MagForce AG holds the European CE certificate (“European Certification”) and thus official approval of NanoTherm® therapy for the treatment of brain tumors in Germany and all member states of the European Union. The effectiveness of NanoTherm® therapy has already been clinically tested on approximately 90 brain tumor patients and in pilot studies on approximately 80 patients with, inter alia, pacreatic, prostate, breast, and esophageal cancer.”

8. Future perspectives

8.1. Cell viability monitoring in MPI–MFH

Another advantage of combining MFH with MPI is that MPI could be used as a real-time sensing tool to monitor the performance of MFH in the future. Most MPI bio-sensing techniques are based on the relaxation of MNPs under an AC magnetic drive field. Magnetic particles undergo two kinds of relaxation: (1) Brownian relaxation, through physical movement of particles; (2) Néel relaxation, through magnetization moment change in individual particles.

Brownian relaxation is highly dependent on the external environment, described by [106], (4) $$\tau =\frac{3\eta v_h}{kT}$$(4) where η is fluid viscosity, v_h is the particle’s hydrodynamic volume, k is the Boltzmann constant, and T is the absolute temperature. Both temperature and external surrounding viscosity can change the Brownian relaxation constant (τ). Temperature sensing based on the relaxation constant of Brownian particles has already been studied [77] and discussed in Section 4.2. As a result, viscosity sensing of Brownian particles is discussed here.

8.1.1. Viscosity sensing using MPI

The presentation of particle relaxation in MPI has been well investigated [68,69,78,107]. Relaxation causes phase delay of magnetization compared to the driving AC magnetic field. Croft et al. [68] demonstrated that the relaxation effect on MPI signal can be modeled by a convolution between the relaxation kernel and the adiabatic MPI PSF, which presents as a blur in the MPI image. Tay et al. [79] at UC Berkeley used pulsed MPI, applying a pulsed square wave AC magnetic field with fast transition (approximately 7 μs), to extract particle relaxation information directly as shown in Figure 7(a). Currently, both the Berkeley scanners and the AWR can measure nanoparticle relaxation. Figure 7(b) shows that our novel pulsed MPI acquisition scheme measures Brownian MPI relaxation time with expected linear scaling with the medium’s viscosity (R² = 0.99). Our preliminary data in Figure 7(c) shows that pulsed MPI relaxometry can also accurately distinguish bound versus unbound MNPs. This is only feasible today by waiting for washout using existing imaging modalities, including MRI, CT, Ultrasound, etc. However, washout times may be impractically slow in many clinically relevant conditions (e.g. binding on capillary walls, or binding in sentinel lymph nodes, etc.).

Figure 7. Berkeley MPI viscosity sensing: (a) Illustration of relaxation measurement. In sinusoidal excitation, relaxation induces a significant lag and blurring of the image. In pulsed MPI, due to the fast transit, the relaxation and the square wave response are well separated in time domain. (Reproduced with permission from Tay et al, IEEE Transactions on Medical Imaging, IEEE [79]) (b) Pulsed MPI relaxation times scale linearly with medium’s viscosity, with ideal R² [108]. (c) Relaxation detects ligand–receptor interaction; streptavidin-coated SPIOs binding to biotinylated albumin [108]. (Reproduced with permission from Hensley, PhD dissertation, University of California, Berkeley 2017 [108]).

8.1.2. Cell viability sensing using Brownian nanoparticles

During hyperthermia, a significant measure of effectiveness is the treatment of a target cell, or cell viability. With current techniques, such as flow cytometry [109], studying cell viability is limited to in vitro experiments. To monitor cell viability during MFH in real time, a new approach must be considered. After MFH treatment, cancer cells go through apoptosis [110], with identifying features including chromatin aggregation, nuclear and cytoplasmic condensation, and partitioning of cytoplasm and nucleus into membrane-bound vesicles [111]. Eventually, these apoptosis bodies, as well as remaining cell fragments, swell and lyse. During this process, there is a change in viscosity, which can be used to sense cell viability. Kuimova et al. [112] observed dramatic intracellular viscosity change during cell death. A significant change of hMSC stiffness during the process of cell death was also observed using an AFM by Nikolaev [113]. To be clear, there is no noninvasive and real time cell viability sensing technique used for MFH right now. MPI viscosity sensing (Figure 7(b) and (c)) could possibly provide a metric for evaluating MFH performance in the future, as target cell response can be tracked during and after treatment. However, for this to be possible, MNPs must be taken up by the cancerous cells. These MPI viscosity measurements can be mapped into viscosity images as demonstrated by Utkur et al. [114].

8.2. Translational efforts to human MPI–MFH system: challenges and opportunities

Recent efforts have shown steps toward clinical scale MPI [104,115]. Graeser et al. [104] demonstrated human-sized MPI for head/brain applications. The only company selling a small animal MPI scanner and a MPI–MFH platform (${\text{Hyper}}^{\text{TM}}$) is Magnetic Insight Inc. The steps of scaling up an MPI–MFH scanner include but are not limited to: designing a gradient coil, designing Tx and Rx coils, and tailoring MNPs. Below, we discuss the hardware challenges and magnetic particle solutions for human scanner.

8.2.1. Hardware challenges

The gradient field is essential to realize clinical localized MFH. This field blocks MNPs outside of the FFR from responding; it only lets particles within the FFR respond to the AC magnetic field. Current small animal MPI scanners use a 7 T/m gradient field, which could provide millimeter spatial resolution for both MPI and MPI–MFH. The high gradient field in an MPI scanner is generated either by two permanent magnets opposing each other or by two high field coils with current flow in different directions [84,116]. While building 7 T/m gradient field for human size scanner is technically challenging, 1 T/m gradient field is feasible and could be realized by flipping the sign of one of the two currents in a $\approx$ 0.43 T whole body Helmholtz MRI magnet. Therefore, it is more affordable than a 0.5 T MRI whole body scanner. The human scanner will have an approximately 7-fold reduction in spatial resolution compared to the small animal scanner and provide a spatial resolution of approximately 10 mm.

A human scanner would also require to scale up the transmit coil and receive coil to fit the size of human body. The sensitivity of both transmit and receive coils scales with B field per unit current, which is approximately proportional to $\frac{1}{R}$ (R is the radius of the coil). Another factor in scaling to a human size scanner is the thermal noise of the receive coil, which is proportional to $\sqrt{R}$ [117]. Therefore, the SNR of the coil scales with $\frac{1}{R^{3/2}}.$ Scaling the coil radius from 3 cm (small animal scanner) to 30 cm (human scanner) will lead to approximately 31.6 times less SNR (assuming that human MPI–MFH keeps the same spatial resolution and magnetic particle concentration as the animal MPI–MFH). Fortunately, MPI is not yet dose-limited in MNP tracers. 2 mM iron oxide is considered safe and current MPI sensitivity is about 1 μM, so there is a safety margin to increase dosage.

8.2.2. MNP solutions

The MNPs used in MPI and MFH follow Langevin function (Figure 2), the magnetization of which saturates at high applied magnetic fields. This is also the reason why a gradient field can localize the particle response in MPI–MFH system. Both spatial resolutions of MPI and MPI–MFH are dependent on the saturation field (H_sat) of the MNP. Smaller H_sat leads to higher spatial resolution. Specifically, the spatial resolution of MPI is the $\Delta H$ of the Langevin derivative plus relaxation effect. Therefore, tailoring MNPs to reach magnetization saturation at smaller field will directly improve spatial resolutions of both MPI and MPI–MFH, reducing the pressure to build stronger gradient coils which will largely reduce the cost of the MPI–MFH system. For example, if the H_sat of the MNPs can be tailored to make the $\Delta H$ of the PSF 10 times smaller, the spatial resolutions of a 1 T/m scanner is approximately 1 mm and 2 mm for MPI imaging and MPI–MFH treatment, respectively.

9. Conclusion

Compared to other treatment methods, the advantages of MFH are apparent: (1) no ionizing radiation, (2) noninvasive, and (3) no depth penetration limitation. It has been successfully used clinically to treat prostate cancer and glioblastoma cancer. Current hyperthermia techniques rely on localized intratumor nanoparticle injection to perform localized heating. However, nanoparticles taken up by the liver and spleen during the clearance process can also be heated by the oscillating magnetic field during the MFH treatment. This side-effect cannot be ignored. Combining MPI with MFH can successfully realize localized heating by confining the magnetization response of the nanoparticle using a strong magnetic gradient field. The heating resolution is gradient-dependent; 7 mm heating resolution was achieved experimentally with 2.35 T/m gradient field. Unlike current imaging-guided methods using CT or MRI which provide off-line imaging references, MPI can provide real time biodistribution information of the MNPs. Implementing advanced sensing techniques of MPI to MFH in the future will advance MFH in making treatment plans, accurate hyperthermia monitoring, and evaluating treatment performance.

Disclosure statement

Professor Steven Conolly is a co-founder of an MPI startup company, Magnetic Insight, and he holds stock in this company. Dr. Daniel Hensley is an employee of Magnetic insight.

Additional information

Funding

We gratefully acknowledge support from NIH grants R01 EB019458 and EB024578, UC TRDRP grant 26IP-0049, M. Cook Chair and the UC Discovery Award. This work was also supported by the National Science Foundation Graduate Research Fellowship Grant No. DGE-1315138 (A.R.R.) and DGE-1842473 (A.R.R.).