Magnetic resonance imaging is a widespread method in clinical diagnostics providing in vivo information on physiological and metabolic changes in tissue. MRI is a non-invasive method because of the low excitation frequency in the radiofrequency range. However, this implies also the low sensitivity of the method, leading to long acquisition times. This can be improved by moving to higher magnetic fields, but on the cost of high prices and large technical demands. Contrast between different tissues is another concern. Contrast agents, as for example Gd-complexes are typically used to improve the contrast. However, in few cases, it induces side effects like allergic reactions and nephrogenic systemic fibrosis (Perazella, 2009; Thomsen, 2008; Kuo et al., 2007). Besides, potential long-term implications arising from the permanent deposition of the Gd-complexes in the brain are currently under discussion (Semelka et al., 2016; Topcuoglu et al., 2020). As an alternative, hyperpolarization methods have been explored to improve sensitivity and contrast for MRI applications. Different strategies have been used so far to hyperpolarize an agent for MRI applications. Optically polarized noble gases as well as para-hydrogen have been used as hyperpolarized molecules for MRI (Salerno et al., 2001; Albert and Balamore, 1998; Limburn et al., 2013; Golman et al., 2001; Duckett and Mewis, 2012; Cavallari et al., 2018). Another method is DNP (dynamic nuclear polarization) where unpaired electron spins excited by microwaves are used to hyperpolarize nuclear spins. Initially this method has been used either at very low magnetic fields or with rapid field cycling (Foster et al., 1998) because microwave radiation cannot penetrate deep into living tissue, or later by hyperpolarizing molecules externally before injection. Nitroxide radicals like TEMPOL are widespread in use for MRI DNP. They have a high Overhauser DNP transfer efficiency to water protons, especially at higher magnetic field strengths (Krummenacker et al., 2012). Furthermore, these radicals can be easily scavenged after the polarization or dissolution step which makes samples biologically compatible and also reduces the shortening of the nuclear T1 relaxation time by the radical (Mieville et al., 2010). However, EPR spectra of nitroxides have three 14N hyperfine lines that hampers a complete saturation of the EPR spectrum at lower radical concentrations. Additionally, for potential medical MRI applications the high radical concentrations needed for Overhauser DNP are not suitable for injection. They can be avoided by immobilization of the nitroxide radicals inside the microwave resonance structure on beads (McCarney and Han, 2008), but the altered dynamics makes the DNP experiments somewhat more complicated and less efficient.

Dissolution DNP is a commercially available hyperpolarization method of liquid substrates for MRI (Olsson et al., 2006). It uses metabolites such as 13C-pyruvate, which are polarized externally to the MRI magnet at very low temperatures (∼ 1 K) by the solid effect. Within 5 min a polarization of 10 % can be achieved (Bornet and Jannin, 2016). After the hyperpolarization procedure is complete, the frozen substrate is rapidly transported to the imager magnet, dissolved in hot liquid, and injected into the body. The typical (dissolving and transfer) time of 5–15 s limits the in vivo usage to 13C and 15N. It is difficult to detect hyperpolarized protons with this approach because of the short relaxation time of water protons compared to the transfer time (Ardenkjaer-Larsen et al., 2014). Therefore, the method is mostly used for the study of cellular metabolism with pyruvate, lactate or other metabolic precursors labelled with low-gyromagnetic ratio nuclei having long relaxation times (Mishkovsky et al., 2012; Jannin et al., 2019). The observation time window is limited by the relaxation time of the hyperpolarized nuclei. Long-lived singlet states might allow to extend this time window (Pileio et al., 2006; Dumez et al., 2017).

Hyperpolarization by the Overhauser mechanism is well suited for physiological solutions under continuous flow conditions, because of the rapid polarization transfer from the unpaired electrons of the radicals to proton spins of the solvent. The polarization transfer in liquids works most efficiently at low magnetic fields. Therefore, first Overhauser DNP experiments for MRI (Guiberteau and Grucker, 1997; Krishna et al., 2002), and proton-electron double-resonance imaging (PEDRI) (Foster et al., 1998) were demonstrated at low magnetic fields of 6–15 mT. This allows irradiation of the paramagnetic radical inside the mouse or rat body in the MRI magnet. Such experiments, however, cannot be performed in clinical MRI scanners at higher magnetic fields (respectively at higher frequencies) due to the strong absorption of microwaves in living tissues, resulting in unwanted heating. A second approach is again to hyperpolarize molecules outside of the target object. This can be done by placing a mw resonator with the radicals into the fringe field of the MRI magnet (Lingwood et al., 2012) or directly inside the MRI magnet close to the target object (Krummenacker et al., 2012). In the first case the hyperpolarization step is done at a lower magnetic field compared to the MRI detection field. This gives a higher polarization transfer efficiency, because the Overhauser effect in liquids is optimal at low magnetic fields. However, the maximal enhancement achieved at the imaging object is downscaled by the ratio between the polarizing and the imaging detection field. For example, in case of an X-band polarizer operating at B0=0.3 T and a 1.5 T MRI scanner this reduction factor is 5-fold. In both cases the fast polarization build-up times allow for continuous delivery of hyperpolarized water, which is essential for monitoring flow and perfusion. In addition, if the hyperpolarization of water reaches a steady state, it eliminates the need for fast MR imaging sequences. In vivo injection of hyperpolarized water safely allows for perfusion imaging in interstitial spaces, localized angiography, and the visualization of brain perfusion because hyperpolarized water freely crosses the blood-brain barrier (Lingwood et al., 2012). Limitations in the accessible area arises from the fast proton spin relaxation.

In our “in-bore” 1.5 T MRI DNP setup there is no such “Boltzmann factor penalty” and a water proton signal enhancement of up to 100 has been reached under static conditions (Krummenacker et al., 2012). On the other hand, the dimensions of fundamental resonators scale down inversely with the mw frequency. Technical challenges are due to the unavoidable high dielectric losses of water in the mw frequency range (Neumann, 1985). These dielectric losses of aqueous samples result in a decrease of the resonators Q-factor, and strong heating of the liquid. The heating can be minimized by applying a resonance mw structure with well separated E and B components of the applied mw field and by placing the sample in a node of the E component. Therefore, cross-section dimensions of the liquid sample inside the resonator have to be much smaller than the mw wavelength. This leads also to very short retention times in case of the flowing liquid. The attainable yield is defined by the highest possible flow rate at a maximum DNP enhancement in such non-stop-flow polarizers. It depends on: (1) the dwell time of water molecules inside the mw polarizer, which should be longer than the proton relaxation time for an optimal DNP build-up, and (2) the mw B1 field strength, which defines the efficiency of saturation of the EPR transitions of the dissolved TEMPOL radicals. The B1 value at the sample depends on the conversion factor c and the quality factor Q of the mw resonance structure, and the applied mw power Pmw (Eq. 1) (Poole, 1967; Rinard and Eaton, 2005). 1B1=cQ⋅Pmw The quality factor Q of the resonator itself depends on the geometry of the mw resonance structure and on sample properties, such as size of the sample and its dielectric constant εr.

Conveniently, cylindrical capillaries made of quartz can be used as liquid sample holders. For this reason, the natural choice of the mw resonator geometry is also cylindrical geometry, such as TE01n type cavities. Such resonators have high Q factors as well as high conversion factors c for frozen solution samples and are commonly used in EPR spectroscopy (for instance, Q = 4000 for an unloaded TE012 resonator at 95 GHz mw frequency has been reported (Smirnov and Smirnova, 2001). This leads to a large B1 field strength, which is advantageous for pulsed EPR applications. The situation is quite different for MRI DNP applications. Here the aqueous sample is in the liquid state, strongly increasing the mw absorption if the sample capillary diameter is not strongly reduced. On the other hand, larger amounts of hyperpolarized liquids are preferable to be able to enhance contrast of as large as possible objects under MRI study.

Fabry–Pérot resonators are mw resonance structures suitable for larger sample volumes, because different from fundamental mode mw cavities, the sample size is only limited in one dimension (direction of the mw excitation) much less than wavelength to avoid absorption of the E field component (Budil and Earle, 2004). Among known quasi-optical mw resonators (Lynch et al., 1988; Prisner et al., 1992; Barnes and Freed, 1997; Milikisiyants et al., 2018; Blok et al., 2004; van Tol et al., 2005; Neugebauer and Barra, 2010) a semi-confocal Fabry–Pérot resonator with the sample placed on the surface of a massive metal (copper) block acting as the plain mirror can best satisfy heat dissipation conditions (Denysenkov and Prisner, 2012).

Here we report on such a new open resonator for Overhauser DNP applications within a whole-body 1.5 T MRI scanner. Particular effort was directed at reaching a uniform (plateau like) microwave field distribution along the flow direction to obtain an efficient polarization transfer to water molecules in the polarizer. The performance of the “in-bore” MRI DNP polarizer equipped with the new resonator was tested and compared to the cylindrical TE013 cavity described before (Denysenkov et al., 2017).

The experimental tests showed that the new resonator has a good mw performance only if the thickness of the water layer is reduced to 50 µm with respect to the value of 80 µm used in the CST simulations. This can be explained by imperfections of the real device geometry caused by fabrication steps, as well as by the higher mw losses (tanδ=1.2) in the water sample at room temperature (Ellison et al., 1996), which was the experimental temperature at the mw characterization of the device with a network analyser with low amount of mw output power (no heating). In the DNP experiments the temperature of the sample is between 65 and 95 ∘C due to the heating by the much higher applied mw power of 10 W. The higher temperature of the water solution in the DNP resonator shortens the rotational and translational correlation times improving the maximum achievable signal enhancements. Additionally this increases the flow rate inside the resonator structure slightly due to fluid expansion. After leaving the mw resonator the temperature of the hyperpolarized solution in the outlet capillary can be brought back to any desired temperature by a heat-sink chiller as described before (Denysenkov et al., 2017) to be physiologically compatible with in vivo experiments.

The optimum flow rate for signal enhancement by DNP is a trade-off between the dwell time of the liquid inside the mw resonator and the flow time through the phantom. The first aspect requires low flow rates to achieve maximum polarization build-up, which is on the timescale of the proton spin relaxation time T1. The second aspect requires high flow rates to reduce the hyperpolarization losses due to proton T1 relaxation determined by the flow time through the phantom. As a consequence of this second point, the highest enhancement is observed at the inlet of the phantom due to the shortest time delay after the DNP polarization process inside the resonator.

The DNP enhancement observed at this point is more than a factor of 2 larger for the new open Fabry–Pérot resonance structure compared to the cylindrical cavity (Fig. 10). The reason is the much larger sample volume inside of the new resonator (25 µL) compared to the cylindrical cavity (6.3 µL), leading to a much longer residence time of the liquid under flow conditions and therefore a larger starting polarization. The achieved B1 field strength with the Fabry–Pérot resonator is only 10 % smaller compared to the cylindrical cavity for the TE013 mode. This is due to the fact that higher Q values can be achieved with the new resonance structure by minimizing the electrical field component of the mw at the sample along the full pathway within the resonator. This cannot be fulfilled as well for the higher mode cylindrical cavity. Similarly, the B1 field in the new resonator is much more homogeneous along the pathway of the sample compared to the cylindrical cavity. All these factors lead to an expected larger DNP enhancement of the liquid at the exit of the new resonator.

The hyperpolarization of the liquid relaxes exponentially back to the thermal Boltzmann equilibrium polarization after leaving the DNP resonator with a time constant given by the proton T1 relaxation time. For the 28 mM TEMPOL solution this time is approximately 0.17 s. Depending on the flow characteristics (laminar, bolus or turbulent), this would lead to a decrease in the polarization of 1/e after approximately 10–15 cm path length inside of the phantom. The experimentally observed decay is much faster due to the chosen gradient echo pulse sequence with a short TR of 110 ms. This time is much shorter than the travel time of the hyperpolarized liquid within the meander-shaped pathway of the phantom. Therefore the hyperpolarized liquid is excited several times by the RF π/2 pulse of the gradient echo sequence. Under our experimental conditions (flip angle of 65∘), each subsequent pulse reduces the hyperpolarization by more than a factor of 2. This can be avoided by applying fast imaging techniques such as echo planar imaging (EPI or SE-EPI) or/and SPatio-temporal ENcoding (SPEN) sequences that are single RF pulse experiments, instead of the multi-pulse GRE protocol used here. However, these sequences are technically more demanding and have to be optimized with respect to contrast and flow artifacts. The fast polarization decay along the flowing liquid due to the short proton spin relaxation time induced by the paramagnetic radicals could be removed by immobilizing the TEMPOL radical inside the DNP resonator (McCarney and Han, 2008; Gajan et al., 2014). This would not only allow the hyperpolarized liquid pathway inside the blood vessels to be observed for a prolonged time and length after the injection point, but also would avoid the exposure of the object to the radical.

The image of the DNP enhanced (bright) signal along the phantom tubing has some dark regions at the positions where the tubing is bent (see Fig. 9b). This feature can originate from some turbulence of the flow at these positions, leading to intra-voxel dephasing, which depends on the degree of turbulence and the direction of the phase- and frequency-encoding gradients with respect to the flow direction (Westbrook et al., 2011). To avoid these additional complications, we compare here the maximum achievable DNP polarization at the inlet point (approximately 5 cm after the resonator). With the new type of resonator, not only larger enhancements, but also a maximum at larger flow rates, was achieved. This is again due to the larger sample volume inside the resonator. Therefore larger amounts and stronger hyperpolarized liquids can be achieved with the new resonator type.

The developed open Fabry–Pérot resonator for Overhauser DNP exhibits an enlarged aqueous sample volume of 24 µL inside the mw resonator, a 4-fold increase with respect to a cylindrical TE013 cavity at 42 GHz. Additionally, the new resonator has a more homogeneous amplitude of the mw B1 field along the sample pathway inside the resonance structure. The design of the resonator and sample geometry also leads to a reduced E1 mw electrical field component along the sample pathway. This leads to a larger Q value of 700 and thus to a similar efficient B1 field amplitude (1.3 G) at the sample compared to a cylindrical TE013 cavity. First MRI DNP experiments with the new resonator demonstrate the possibility of increasing the flow rate of hyperpolarized solutions significantly (4-fold) as well as of reaching higher DNP enhancements (2-fold). The obtained improvement is encouraging and might together with using immobilized TEMPO radicals be a promising approach for in vivo MRI angiography applications on small animals in the future.