Merging Preclinical EPR Tomography with other Imaging Techniques

This paper presents a survey of electron paramagnetic resonance (EPR) image registration. Image registration is the process of overlaying images (two or more) of the same scene taken at different times, from different viewpoints and/or different techniques. EPR-imaging (EPRI) techniques belong to the functional-imaging modalities and therefore suffer from a lack of anatomical reference which is mandatory in preclinical imaging. For this reason, it is necessary to merging EPR images with other modalities which allow for obtaining anatomy images. Methodological analysis and review of the literature were done, providing a summary for developing a good foundation for research study in this field which is crucial in understanding the existing levels of knowledge. Out of these considerations, the aim of this paper is to enhance the scientific community’s understanding of the current status of research in EPR preclinical image registration and also communicate to them the contribution of this research in the field of image processing.


Introduction
The application of electron paramagnetic resonance (EPR) spectroscopy and imaging to human subjects has been made possible by recent technical and scientific developments. The wide range of electron paramagnetic resonanceimaging (EPRI) applications enabling mapping live tissue microenvironment parameters such as redox state, pH, thiol concentration, and especially oxygen partial pressure (pO 2 ). This makes EPRI an attractive functional-imaging modality for medical applications. However, insufficient anatomical details provided by EPRI, necessitates the use of other anatomic-imaging references for accurate interpretation of the EPR images, especially in an in vivo setting. The question arises how to merge EPRI with tissue structureimaging modalities. This is important in the context of future human applications and preclinical studies.
Anatomic reference can be provided by nuclear magnetic resonance imaging (MRI), X-ray computed tomographic imaging (CT), or ultrasound imager. Several different approaches to correlate anatomical information with EPRI have been developed over the years. These include: building dual-modality imagers where both MRI and EPRI is performed on the same animal [10], applying image registration procedure [11], choosing anatomically referenced region of interest (ROI) by using loop resonator [12], modulated field gradient EPRI [13] allowing to image only a chosen part of the body, and overlaying EPR images with anatomical atlas. For ex vivo measurements such as bioptates, excisions, and histological sections, photographic images of the samples can be used [14][15][16]. All of the above solutions allow for merging functional EPR images with an anatomical reference. imaged object. Even though current imagers allow to obtain high spatial resolution images, they still only provide information about the volume in which spin probe is distributed, without accurate tissue borders or discernment between different tissues. For example, during tumor imaging the distribution of the spin probe is not very different between tumor and muscle in EPRI spatial images, while in MRI the tissue structures are clearly visible (Fig. 1).
Therefore, complementary techniques for anatomical reference need to be implemented together with EPRI. To meet this challenge the most often used solutions are: (i) developing hybrid systems for simultaneous imaging using both nuclear and electron magnetic resonance without having to move the animal between imagers; or (ii) image co-registration, which allows for fusion of the information from different images of the same object, regardless of their modality. Both approaches have opened a new opportunity window for biomedical research; finally, (iii) even if neither of above may be applied, there is a possibility of implying internal anatomy or structure from the a priori knowledge.

EPRI-MRI Hybrid Systems and PEDRI
Similar physical and technical foundation of the methods stimulated inventors towards developing hybrid systems that allow to conduct MRI and EPRI experiment in the same device. Such approach was first reported by the Zweier group [17]  2010 the same group built a dual-frequency resonator for EPR/MRI co-imaging, to avoid errors in EPR and MRI coimaging registration caused by inadvertent animal movement [20]. Recently, several studies have been published demonstrating that the hybrid approach has become a mature technique, bringing novel information in various settings. For example, Caia et al. have applied EPR/MRI coimaging system for the mapping of in vivo redox state both in control and cigarette smoke-exposed mice [21]. Another approach combining MRI end EPR was to employ EPRrelated microwave irradiation with low-field magnetic resonance imaging so that the electron spin polarization is transferred to nearby protons, resulting in stronger signals. This approach is called proton-electron double-resonance imaging (PEDRI), also known as Overhauser MRI (OMRI) or a liquid-state dynamic nuclear polarization (DNP MRI). Imaging of PEDRI is effectively an indirect form of EPRI in the sense that enhancement of proton signals is proportional to the concentration of spins in the paramagnetic system and the degree of saturation of the EPR system. In PEDRI image, regions of the sample which containing free radicals have much higher intensity due to the Overhauser-enhanced proton nuclear magnetic resonance signal, revealing the location of the free radicals [22][23][24]. A significant impact in this field was given by Prof. Lurie and Prof. Zweier groups, by developing PEDRI of free radicals using a clinical MRI system [10] and then upgrading it successfully into the coregistered system. This approach allowed for the MRI/EPR image registration of living mice, so that the whole procedure of setup and imaging took~40 min [17,19]. This way EPRI has been linked with MRI, and thus, functional images of oxygenation or redox state obtained by EPR have been fused with high-resolution anatomical images provided by MRI. Li et al. used a PEDRI system for rapid imaging and evaluation of the in vivo distribution and clearance of triarylmethyl radical in mice [25]. In 2014 Samouilov et al. have used an advanced nitroxide probe to image extracellular tumor pH using double-resonance imaging, providing an analytical approach for spatially resolved noninvasive pH monitoring, in vivo [26]. An exhaustive review about in vivo application of PEDRI was delivered by Kishimoto et al. [27]. In Table 1 significant studies where hybrid systems EPR/NMR and PEDRI were used for in vivo applications are displayed and an example of in vivo application of the hybrid system, showing the distribution of charcoal in the body of a mouse is shown in Fig. 2 [19].
EPR/MRI hybrid imagers and PEDRI systems make it possible to obtain both anatomical and functional images provided by these two magnetic resonance techniques without moving the specimen. The benefits of such solution fulfill the requirement of anatomical reference for EPRI maps from a single experiment in relatively short timescales, without changing the position of the imaged animal. So far, however, this approach is limited to two magnetic resonance methods. To involve more imaging techniques where physical basics are not as similar as EPR/ MRI, such as CT, PET, SPECT or ultrasound, a correct fusion of multiple images from different methods has to be performed. The process of accurate overlapping of the images from separate imaging techniques is referred to as image registration and was initially developed for clinical applications [28,29].

Image Registration
Image registration is the process of finding correspondence between all points in at least two images of the same object [11]. Images may be two or more dimensional, be obtained from different modalities, have different resolutions and orientations. For the images which do not possess sufficient common information, like in the case of EPR and MRI images, artificially added common points of reference, typically referred to as fiducials, are used. A key element in image registration lies in the proper selection of fiducials markers. Fiducials markers are usually rigid objects placed in the imaging field of view in such way that they are fixed in relation to the imaging region. Four nonplanar points in space which do not belong to any symmetry group establish a unique definition of linear coordinate system which for registration procedure ensures the unambiguous location. Moreover, these markers have to be visible in all acquired images from each modality used. In the case of EPRI, thin tubes filled with the radical-containing solution can be used. For registration reference, the radical should exhibit a single, narrow EPR line, which is resulting in high-quality EPR images. For MRI a similar approach can be used with fiducials filled with water. Fiducials filled with a lowconcentration radical solution can be visible both in EPRI and MRI. Other modalities as CT or micro-CT typically use heavy-metal containing markers, however, glass tubes can also give sufficient contrast. Thus, markers seen by EPRI, MRI, and CT may consist of thin glass tubes placed close to the imaged object, filled with aqueous solution of a paramagnetic spin probe.
The fixing of the fiducials to the imaged object and ensuring the same position in two (or more) imaging modalities is completed by two methods: (i) building a single-use cast or a mold using a non-paramagnetic material which immobilizes the imaged object and the fiducials; (ii) designing a multiple-use cradle or bed for the animal, with built-in position for fiducials. Both approaches have their advantages and limitations. Building a cast is useful when only a part of the animal body is going to be imaged, e.g., the leg with the tumor. The custom cast for the animal may be prepared very quickly, in such way that it adequately fills the resonator and allows positioning the object ROI and the fiducials in the center of the resonator. When the imaging device changes, an animal with the cast intact is moved between instruments. The second approach, using a special bed or cradle-containing fiducials adapted to hold animals may be more advantageous for a whole-body experiment. If the cradle is compatible with all imaging instruments, this approach might be better for the animal.
A crucial role in the preclinical EPR-imaging research, where anatomy images were obtained using MRI, was played by Halpern's group. Using vinyl polysiloxane cast material or 3D-printed animal cradles Halpern's group had a lot of success in the correlation of in vivo pO 2 images from mice, rats, and rabbits not only with MR images but also with CT and PET modalities. Vinyl polysiloxane dental material was shown to be a convenient material for immobilization of animal and fiducials close to an animal leg. It has been successfully used to multimodality-imaging combining EPRI with MRI, (Fig. 3, [30]). Having established the procedure of registration of the EPRI with MRI, in all the subsequent studies, EPRI hypoxia maps of tumors were supported by accurate tumor anatomy images obtained from MRI. For example, they have demonstrated that EPR image hypoxic fraction separates the population of fibrosarcomas that are cured from those that fail after a single radiation dose in in vivo model [31]. Similarly, it was shown that tumor hypoxia measured by EPRI is highly predictive for their response to radiotherapy. Data showed that low hypoxic tumors treated to nominal TCD 50 were ca. three times more controlled than poorer oxygenated tumors [32]. Haney et al. reported that a quantitative absolute oxygen measurement correlated with tissue perfusion as determined by DCE MRI, in vivo [33]. This image-guided approach for antivascular therapy may help to identify regions of good and poor response for treatment [33]. In 2010 Epel et al. presented methodologies for EPR/MRI image registration in oxygen imaging in the rabbit leg bearing a VX-2 tumor (Fig. 1) [34]. This study paved the way for the application of EPR imaging to study animals larger than mice and rats. The same procedure of image registration was applied to assess the thiol redox state, specifically the estimation of intracellular thiol concentration in vivo. Epel et al. used (RS)rapid scan EPR imaging to obtain 3D EPR images in very short time,~30 s [8,9]. Changes in EPR line of PxSSPx spine probe sensitive to intracellular thiol concentration were acquired on physiologically relevant timescales and illustrated the enabling capability of RS EPR imaging.
Supporting EPRI by other imaging methods, especially MRI in both hybrid systems and with image registration process provides important and unique information about tumor physiology and prospects for a new range of disease treatments. However, MRI is not the only modality providing the necessary anatomical information. Beziere et al. in 2012 demonstrated image registration of EPRI with micro-CT for the study of the knee joint in the mouse. EPRI was used to measure the intra-articular oxidative stress involved in various pathological situations, including osteoarthritis, and rheumatoid arthritis in the knee joint (Fig. 4, [35]).
Another image registration application taking EPRI into account, more specifically redox state of mice brain, was exhaustively investigated by the Hirata's group. One example is a correlation of brain MRI with redox state in a brain-diseased mouse model using nitroxides as a spin probe [36]. Further studies relating the use of EPR imaging in vivo with image registration are summarized in Table 2.

Registration Using Features of the Images
Functional images, such as oxygen images may not exhibit sufficient features for precise registration. Therefore, the application of algorithmic feature-based registration such as entropy-based registration may not be practical. Similar functional images, however, may exhibit sufficient common information. In addition to the information provided by the images, one could use a priori information such as animal anatomic details, organ localization, approximate orientations of the object in the imagers, etc. Even if such Fig. 4 In vivo image registration micro-CT and EPR of mouse knee joint after intra-articular injection of TAM. Frontal (upper) and medial (lower) views of the raw EPR image of the mouse knee joint and of two glass capillaries containing a TAM solution that were placed on each side of the mouse knee joint to locate it. b Anatomic x-ray micro-CT image of the same mouse knee and capillaries. c Spatial registration, and fusion of the raw EPR and micro-CT images using capillaries as fiducial points. d Molecular EPR image of the mouse knee joint after mathematical reconstruction of the image taking into account TAM disappearance during data acquisition. e Spatial registration of image D to the micro-CT anatomic image (b) of the same mouse knee. On b,1 5 patella; 2 5 femur; 3 5 joint cavity; 4 5 tibia; 5 5 fibula; 6 5 reference capillaries. A small volume of TAM was detected away from the bone on the inner side of the knee (see d and e). This could be a small drop of radical located on the skin, where the needle was inserted (adapted from [35])  [48] 4D (pO 2 maps) Ox063 OxyLite Tumor Hyodo et al. [49] 2D (spin probe distribution) 3-CP MRI Mouse heart, liver, and kidneys Matsumoto et al. [50] 2D (redox status) 3-CP MRI Tumor Elas et al. [31] 4D (pO 2 maps) Ox063 MRI Tumor Matsumoto et al. [51] 4D (pO 2 maps) TAM MRI Tumor Haney et al. [33] 4D (pO 2 maps) Ox063 DCE-MRI Tumor Epel et al. [34] 4D (pO 2 maps) Ox063 MRI Tumor (rabbit) Elas et al. [52] 4D (pO 2 maps) Ox063 Biopsy Needle Tumor Fuji et al. [36] 3D (redox status) HMP MRI Mouse head Matsumoto et al. [53] 4D (pO 2 maps) TAM 13C-MRI Tumor Elas et al. [32] 4D (pO 2 maps) Ox063 MRI Tumor Takakusagi et al. [54] (pO 2 maps) Ox063 MRI Tumor Radler et al. [55] 3D (spin probe distribution) Nitroxides c , Ox063 MRI Tumor Neveu et al. [56] 3D (redox status) Ox063 13C-MRI, PET, CT Tumor Epel et al. [9] 3D (intracellular thiol concentration maps) PxSSPx d MRI Tumor a Triarylmethyl radicals b 3-Carbamoyl-proxyl c cis-3,4-Di(acetoxymethoxycarbonyl)-2,2,5,5-tetramethyl-1-pyrrolidinyloxyl, cis-3,4-dicarboxy-2,2,5,5-tetra-methyl-1-pyrroldinyloxyl d disulfide-dinitroxide registration is rather unprecise, it is worthwhile to present EPRI images in anatomical or structural context. Many significant biomedical studies have been performed using approximate registrations. E.g., Williams et al. have shown consistency between EPRI pO 2 maps of tumors and BOLD MRI [37]. Sonta et al. demonstrated an increased oxidative stress in the kidneys of diabetic mice by comparison of EPRI studies with immunohistological staining [38]. Further important work where EPRI pO 2 imaging was compared with other hypoxia-imaging techniques such as PET, OxyLite, or 19 F-MRI were performed by Tran et al. in 2012 and showed a good correlation between the tested methods [39]. Oxidative stress study performed ex vivo in human biopsies was reported by Gustafsson et al. They presented the distribution of oxidative stress of atherosclerotic plaques mapped using EPRI of ROS sensitive spine probe and compared it against histopathology images [14]. Significant examples where EPRI has been compared with other imaging techniques without utilizing hybrid systems or the registration procedure are shown in Table 3.

Conclusions
Merging other imaging techniques with EPR imaging seems to be beneficial both for human and preclinical studies. Functional information from EPR imaging supplemented with anatomical knowledge delivers information about oxygenation, redox state, pH, or thiol concentration to be obtained from a chosen ROI. It is crucial that such information is precisely located in multidimensional anatomical space, so that all its aspects-spatial, spectral, and temporal can be correctly interpreted. Thanks to the continuous development of EPR methods, software with a user-friendly interface, combining EPR devices with other systems and the synthesis of advanced spine probes, these solutions enable obtaining many significant results in both human and preclinical biomedical studies.