Medical Imaging of Microrobots: Towards In Vivo Applications

Medical microrobots (MRs) have been demonstrated for a variety of non-invasive biomedical applications, such as tissue engineering, drug delivery, and assisted fertilization, among others. However, most of these demonstrations have been carried out in in vitro settings and under optical microscopy, being significantly different from the clinical practice. Thus, medical imaging techniques are required for localizing and tracking such tiny therapeutic machines when used in medical-relevant applications. This review aims at analyzing the state of the art of microrobots imaging by critically discussing the potentialities and limitations of the techniques employed in this field. Moreover, the physics and the working principle behind each analyzed imaging strategy, the spatiotemporal resolution, and the penetration depth are thoroughly discussed. The paper deals with the suitability of each imaging technique for tracking single or swarms of MRs and discusses the scenarios where contrast or imaging agent’s inclusion is required, either to absorb, emit or reflect a determined physical signal detected by an external system. Finally, the review highlights the existing challenges and perspective solutions which could be promising for future in vivo

Scientists around the world have aimed for a long time at developing miniaturized robots that can be controlled inside the human body to aid doctors in identifying and treating diseases. The visionary idea of performing non-invasive medical interventions by using tiny machines, postulated by Albert Hibbs and Richard Feynman back in 1959, 1 is not considered science fiction anymore, and is becoming a reality. These miniaturized robots, here called MRs are tiny structures with simple geometries such as rods, tubes, spheres, and helices, which can be propelled by chemical reactions, 2-9 physical fields, 10-21 motile cells or microorganism inclusion, [22][23][24][25][26][27][28] and which can perform several functions, 27,[29][30][31][32][33][34][35] such as minimal invasive surgery, 13,36,37 sensing, [38][39][40] targeted drug delivery or diagnosis. 30,41 Precise control over MRs locomotion and function is necessary to enable target reaching and efficient task performance; however, fine control is possible only if information on MRs position and configuration is available. Few examples of MRs in living mice and small animals have been reported for cancer and bacterial infection treatments, [42][43][44][45][46] but the way ahead to the clinical practice and to high therapeutic efficiency is still long. In fact, despite a rapid development in the last decade, most of the MR applications demonstrated till now have been enabled by the use of optical microscopy in in vitro conditions and translating such approaches to an in vivo environment towards real medical applications is not straightforward. 47 Medical imaging techniques should be employed to visualize the physiological environment, including potential vascular routes and the target site for therapy, and for assessing the position of MRs in biological environment. In this framework, a wide plethora of medical imaging techniques are available.
However, employing standard medical imaging techniques for microrobots localization is not trivial and many challenges still need to be faced in the attempt to identify the best compromise between spatiotemporal resolution, field of view (FoV), penetration depth and level of invasiveness associated with the imaging technique. The suitability of each imaging technique depends also on the working body districts and on the possible employment of swarm control strategies. For example, organs such as the eye or the skin can be analyzed using imaging techniques with low penetration depths. 48,49 In contrast, applications such as cancer treatment 50 in internal organs or assisted in vivo fertilization, 12,22 just to mention some examples, require deep tissue imaging.
In this paper, we critically review the state of the art of MRs imaging in phantom, ex vivo, and in vivo, including both conventional and hybrid imaging techniques. A particular focus is given to the physics behind each technique in the attempt to unveil the opportunities and limitations of each imaging strategy with the aim to provide the readers with useful indications and guidelines to steer future research in the field of microrobotics. However, understanding the basic principles behind the different imaging strategies and analyzing their potentialities to resolve microscale structures in biological environment, will also be of interest for a wider healthcare research community interested in performing deep tissue imaging with high spatiotemporal resolution for investigating biological and synthetic mechanisms such as microcirculation, vascularization of implanted materials, cell migration, microrheology or any dynamic phenomena occurring at the micro and nanoscale.
Among "conventional techniques", optical, ultrasound, magnetic and radiation-based imaging methods are described, in particular when used to track medical MRs. Likewise, "hybrid techniques" such as photoacoustic (PA), and magnetomotive ultrasound (MMUS) imaging are reviewed. The review also deals with the suitability of each technique for single microrobot or swarm tracking and discusses the scenarios where contrast or imaging agents are required. Finally, some unexplored concepts combining different excitation sources (e.g. light, magnetic field, ultrasound) as well as different detection principles (e.g. acousto-mechanics, light reflection, wavefront shaping) are put into perspective, elucidating different means of imaging that can outperform current techniques in the future for MR-assisted theranostic applications.

CONVENTIONAL IMAGING TECHNIQUES
All imaging techniques share the same basic principle: a physical signal passes through the body/area under diagnosis and its interaction with the tissue causes transmission or reflection of the radiation which is then captured by a detector array and processed into an image pattern. 51 Different energy sources can be exploited ranging from electromagnetic fields (including light and high energy radiations) to ultrasound.
In this review paper, clinically established and widespread imaging techniques such as magnetic resonance imaging (MRI), ultrasound (US), optical and ionizing radiation-based techniques are defined as "conventional techniques" although they are continuously improving in terms of technical specifications for future potential applications. Based on the underlying physics, we classify the conventional techniques into (i) magnetic field, (ii) ultrasound, (iii) optical and (iv) ionizing radiation-based ones ( Figure 1A). MRI and magnetic particle imaging (MPI) are described as magnetic field-based techniques whereas pulse-echo techniques, such as B-mode and Doppler-mode imaging, are included in the US-based techniques. Optical-based techniques to visualize MRs are classified into two groups, namely reflection-based imaging (RI) and fluorescence-based imaging (FI). Last, imaging techniques employing high energy radiation will be analyzed by describing in detail Computed Tomography (CT), X-rays, fluoroscopy, Positron Emission Tomography (PET) and Single Photon Emission Computed Tomography (SPECT) principles and employment in MRs localization and tracking. Although various other imaging techniques may be available, we discuss the most promising candidates for the localization and tracking of MRs in terms of three key parameters, namely spatial and temporal resolution as well as penetration depth ( Figure 1B). Spatial resolution is the parameter which defines the minimal object size guaranteeing imaging or alternatively, which is the minimal distance possible between two objects to be discriminated from each other. Temporal resolution is the minimal time required 6 to complete the data acquisition needed to reconstruct one complete image of the object. Penetration depth is the measure of how deep the probing signal can penetrate into tissues and can be defined as the distance where the primary intensity is reduced to l/e (i.e. 37%) of the initial transmitted intensity. Additionally, another crucial parameter is the contrast resolution (or simply contrast), defined as the ability to distinguish between different levels of intensity in an image. Fluorescence Imaging (FI); X-ray Imaging (X-ray); Positron Emission Tomography (PET); Single Photon Emission Computed Tomography (SPECT); Photoacoustic (PA) Imaging; Magnetomotive Ultrasound (MMUS); Optical Coherence Tomography (OCT). Dark gray region indicates the ideal scenario for spatial and temporal resolution for tracking medical MRs, and the ideal penetration depth is marked as yellow-orange color.

Magnetic Field-based Techniques
Magnetic field-based techniques are well-established and widely used in biomedical imaging to investigate the anatomy, physiology of the body and to detect pathologies including tumors, inflammation, strokes or abnormalities in the heart or blood vessels. Magnetic fields in the range of few Teslas can propagate in deep tissues and interact as non-destructive radio waves. These interaction phenomena range from electrodynamic exchanges to the generation of mechanical forces and torques, and can be translated into detectable electrical signals. Magnetic-based techniques allow functional (by using magnetic tracers) and anatomical imaging (thanks to the spin of hydrogen atoms abundant in soft tissues), as in the case of MRI, but also measuring the spatial distribution of specific materials, as for the scenario of MPI. These two techniques typically rely on expensive coil-based equipment and call for dedicated powering sources and cooling systems that significantly increase the complexity of the imaging apparatus. Two key features of magnetic techniques lie in the high penetration depth, especially in MRI due to the low attenuation of magnetic fields across tissues, and in the possibility to combine imaging and manipulation of MRs with the same apparatus.

Magnetic Resonance Imaging
MRI is particularly suitable to acquire 3D anatomical images of soft tissues and is used for this purpose in clinical practice. This can be achieved with a high spatial and temporal resolution with and without the use of specific contrast agents. Progress in MRI research has been tremendous in recent years, feedback control units have been significantly improved, and it has become one of the most appealing methods not only for medical imaging, but also for targeted minimally invasive therapies. 25,52,53 MRI is based on the principle of atomic spin relaxation. The term relaxation is used to describe the process by which a nuclear spin returns to thermal equilibrium after absorbing radiofrequency (RF) energy. 54 Three magnetic field stimuli are combined in MRI, namely a static magnetic field, electromagnetic RF signals and magnetic field gradients for localization ( Figure   2A). 55 In MRI the nuclear spins of hydrogen atoms, naturally abundant in tissues and biological materials, align under the effect of high intensity static magnetic fields (typically between 1.5 and 3 T). Pulses of radio waves are employed to excite the nuclear spin energy transition, which tend to realign with the static field in the process of relaxation. There are two principal relaxation components, i.e. longitudinal and transverse, related to the time constants T1 and T2, respectively ( Figure 2B). Magnetic field gradients allow for the localization of these signals in space.
By varying the parameters of the pulse sequence, different contrasts may be generated between tissues (or non-biological objects as MRs can be) according to the relaxation properties of the hydrogen atoms therein. Based on its operating principle and on the capability of the technique to detect almost any kind of material, MRI can be considered as a label-free technique not calling for the employment of dedicated contrast or imaging agents. However, when considering micro-and nanoscale objects, magnetic materials could be employed to increase the signal intensity and enhance image quality. Superparamagnetic and paramagnetic agents, based on gadolinium ion complexes or on iron oxide nanoparticles (NPs), are particularly important as MRI contrast agents due to their high relaxivity (i.e. relaxation rate as a function of concentration). Magnetic materials subjected to the static magnetic field of a clinical MRI scanner produce a locally perturbing dipolar field with a shortening of the transverse relaxation time T2. 56 Based on the strength of the magnetic object, these distortions can be sufficiently large to detect MRs with overall dimensions smaller than the spatial resolution of any clinical imaging platform. Empirical observations suggest that an iron oxide particle distorts the magnetic field distribution over a region corresponding to about 50 times the particle size. 57 In this way, given that the typical size of a voxel in a standard 3T MRI scanner is about 500 μm, it is reasonable to say that single microrobot imaging could be foreseen for agents at least 10 μm in size (if magnetic properties are suitable enough) whereas swarm imaging appears more realistic for smaller or weaker magnetic microrobots.
In the early works by Martel and co-workers, [58][59][60] MRI was one of the initial clinical imaging techniques to be employed in the micro/nanorobotics scenario. In addition, MRI scanners produce strong magnetic fields and gradients which can be exploited for navigation of magnetic microrobots. [61][62][63] Martel and co-workers used MRI to image and steer magnetotactic bacteria and magnetic beads in vitro and in vivo (e.g. swine) across the vasculature (Figure 2C-E). 59,64,65 Interesting works from this group dealt with thermoresponsive hydrogel-based magnetic soft microrobots for targeted drug delivery. 66 In most cases, MRI was employed to image swarms of MRs (few µm in size each). Zhang and co-workers successfully employed MRI to track in rodent stomachs a swarm of magnetic helical microswimmers, obtained from Spirulina microalgae via a facile dip-coating process in magnetite (Fe3O4) suspensions. 43 They demonstrated the higher penetration depth enabled by MRI in comparison with fluorescence-based imaging. However, MRI imaging of single agents has been demonstrated only at the millimeter scale. For instance, Fatikow and co-workers exploited imaging artifacts produced by a single mm-scale ferromagnetic object for MRI-enabled closed loop control (tracking and navigation). Path planning algorithms based on MRI data were integrated with a tracking module providing feedback on the position of the ferromagnetic object. 67 Reported works using MRI imaging to track MRs are summarized in Table   1.
Main challenges of medical MRI in the field of microrobotics are associated with spatial and temporal resolution. The spatial resolution is crucial for single microrobot visualization and for detecting events like target reaching, shape transformation or activation of drug delivery mechanisms. In MRI, the spatial resolution is directly proportional to the sampling rate of the relaxation signal, and inversely to FoV and slice thickness. Spatial resolution can also be improved by increasing the acquisition time, but that comes at a reduced framerate with inferior temporal resolution. In a previous study, the detection of a limited number of magnetic microrobots by MRI was achieved when increasing the acquisition time from seconds to few minutes. 68 For any specific application, an optimal compromise between spatial and temporal resolution needs to be identified.
Furthermore, when combining MR-based navigation and MRI, we should consider that different magnetic field signals are required for the two purposes. This calls for the necessity of implementing time-dependent multiplexed sequences in which navigation and imaging alternate and control relies also on pre-operative path planning. Possible delays and navigation instabilities produced by the switching between imaging and actuation could be minimized by appropriate processing algorithms or by employing multimodal imaging strategies, e.g. combining MRI with X-ray CT imaging. 69 Alternatively, off-line imaging is often used to assess the targeting efficacy of microscale agents following the navigation phase and appears particularly interesting for MRs-based applications with low dynamics. Publishers guidelines. 65 Copyright SAGEPub (2009). C) Unfiltered vasculature images (without any microrobot placed in the vessel). D) Distorted MRI image created by a 1.5 mm chrome steel sphere and E) Tracking of the same sphere over a pre-acquired X-ray image while being controlled at 24 Hz in the carotid artery of a living pig.

Magnetic Particle Imaging
MPI was proposed in 2001 by Bernhard Gleich and Jürgen Weizenecker. 70 The technique allows quantifying the presence of paramagnetic material in the FoV upon saturation of their magnetization. MPI cannot image tissue but can assesses the 3D distribution of paramagnetic material (e.g. NPs). To build an MPI scanner, two permanent magnets are arranged in a Maxwell configuration (i.e. opposing each other with identical poles) so that the magnetic field intensity is null in the central region of the arrangement, named the field-free point (FFP). Electromagnetic drive coils produce a time-varying magnetic field in the imaging area and dedicated pick-up coils detect changes in the NPs magnetization. Magnetization saturation of particles produces non-linear effects that generate harmonics which intensity is proportional to the particle's concentration. The static gradient field allows for spatial encoding and for detecting the origin of the signal. More specifically, at the FFP the particles magnetization is free to follow the drive field whereas outside the FFP the particles are saturated by the high-intensity static field, thus not producing any detectable signal ( Figure 3A).
The FFP can be moved in space by introducing additional time-varying focus fields through the drive coils. This enables 3D volume scans with high temporal resolution within spherical workspaces of 10-20 cm in diameter. More specifically, MPI provides the capability of covering a volume similar to that of the entire heart or brain with more than forty 3D acquisitions per second.
The spatial resolution of MPI increases with the intensity of the magnetic field gradient and with the particle's magnetic susceptibility. It lies in the range of one to few millimeters for currently developed MPI scanners. 71 In the past decade, researchers have reported promising results regarding the use of MPI for the visualization and navigation of small magnetic devices. The high-intensity field gradients generated within the scanner workspace allow to pull and actuate magnetic objects. In 2016, Gleich and coworkers demonstrated 1-D and 2-D closed-loop control and imaging of a millimeter-scale magnetic device using a prototype MPI setup. 72 Analogously to MRI, imaging and actuation require different drive field sequences in MPI. For this reason, controlled navigation is performed in a quasi-simultaneous manner by time-dependent multiplexed sequences in which actuation and imaging alternate. For this reason, even if imaging can be performed at very high frame rates, the achievable control frequency is limited by the time required by magnetic actuation.
In 2018, researchers from the same group developed a clinical scale system which was successfully used for 3D actuation of a centimeter-scale magnetic drill first in a phantom and then in ex vivo tissues. 73 This study highlighted how scaling up the workspace implies a reduction of the available spatial resolution, as a result of smaller magnetic field gradients. In 2018, Gleich and coworkers developed a human-sized MPI platform for brain applications ( Figure 3B). 74 However, the achievable spatial resolution was about 10 mm, which is not suitable for microrobotic applications ( Figure 3C).
Smart materials can improve the spatial resolution of MPI. In 2019, Bakenecker and coworkers reported a 3D-printed millimeter-scale helical swimmer (2 mm in length) and navigated it through a vessel phantom with an MPI scanner ( Figure 3D). 75 The swimmer was coated with paramagnetic NPs (130 nm in diameter) to enable both imaging and navigation. With this setup, it was possible to obtain images of the navigated propeller (sub-millimeter resolution) at 10 frames per second (fps). In 2020, Griese and coworkers performed a comparative study in which swarms of different NPs were navigated through a vessel phantom in bifurcation flow by an MPI scanner. 76 They found dextran-coated iron oxide NPs with 250 nm diameter to provide the best compromise in terms of magnetic manipulability and imaging performance.
Overall, fast localization and strong forces due to the high field gradient render an MPI system as a good platform for image-guided steering of magnetic devices. Nevertheless, some limitations still hamper the translation of this technology into clinically-relevant scenarios. On one hand, more research efforts are needed to scale up the platforms and to enable human-sized workspaces, while guaranteeing good imaging performances. On the other hand, the spatial resolution seeks for improvement (a few millimeters with current platforms). Innovative solutions are required to progress in this direction, both in the design of high gradient selection fields and in the development of suitable materials with steeper magnetization characteristics. Adopted with permission. 75 Copyright Elsevier (2019).

Ultrasound-based Techniques
Amongst conventional techniques, US-based imaging stands as one of the most promising solutions for providing real-time feedback on microrobot position in deep tissue. It is a mature technology regularly used in clinical settings for diagnostic purposes. Compared to other conventional techniques, US combines high spatial and temporal resolution with minimum adverse effects on tissues and lower equipment cost. In US imaging, pulses of pressure waves are emitted by a piezoelectric transducer. When the waves encounter acoustic impedance discontinuities, they are partially scattered back towards the source. 77 The backscattered echo signal is registered by the transducer and can be processed to extrapolate information regarding the whole journey of the transmitted wave. Several US beams can be independently shot and registered back by an array of piezoelectric elements (US probe) to acquire adjacent scan lines which are then combined to reconstruct a 2D tomographic image, also known as a frame. In this review, we are going to discuss two different established US imaging modalities, namely B-mode and Doppler, which look particularly suitable for MR tracking.

B-mode Ultrasound Imaging
B-mode US is a pulse-echo imaging modality in which a pulse wave is transmitted by the US probe and the echoes generated by the interaction of the wave with the propagating media are processed to reconstruct an image of the insonated region (i.e. the region exposed to ultrasound). Assuming to transmit a sinusoidal pulse along a single scan line, the detected echo * ( ) can be expressed as: where ( ) is the amplitude of the signal, is the carrier frequency of the transmitted pulse and ( ) is the phase. The instantaneous amplitude ( ), also known as the envelope of the signal, carries information about the pressure intensity that is locally backscattered by all objects along the path of the scan line. In brightness-mode (B-mode) imaging, ( ) is converted into brightness levels to create a contrast image of the object ( Figure 4A). When propagating in human tissues, US waves are strongly absorbed by bones (producing acoustic shadows) and highly reflected by air sacs. Due to its strong echogenicity (i.e. the property of backscattering US waves), the air is generally used as a contrast agent, for instance in the form of microbubbles. These are micronsized spheres consisting of a gas core surrounded by a lipid shell which act as echo-enhancers. 78 In B-mode imaging, the temporal resolution depends on the speed of sound in the medium (about 1500 m/s in soft tissues) 79 and on the FoV. Typical frame rates are in the order of 50-100 fps, while greater rates are achieved by plane wave imaging (>1000 fps). 80 The spatial resolution is defined by two parameters, namely axial and lateral resolution. The lateral resolution depends on the geometry of the US probe since it is related to the distance between the piezoelectric elements of the probe array. Axial resolution is related to the wavelength of the transmitted pulse (typically 100-500µm) and thus to both the characteristics of the transducer (center frequency) and the properties of the medium (speed of sound). Due to wave diffraction, objects with characteristic dimensions comparable to the US wavelength are not well resolved. Ideally, axial resolution is defined by one half of the transmitted pulse wavelength. However, as a result of the combined effect of the physical processes involved (i.e. scattering and diffraction) and of the reconstruction procedure (e.g. imaging artifacts), the axial resolution of a B-mode image is usually worse than the ideal value. 81 Figure 4B). 84 To localize such a small microjet in the imaging plane, they exploited the echogenicity (i.e. characteristics of large wave reflection) of gas bubble trails produced by catalysis-based propulsion. However, the average position tracking error reported in this case was 250.7 ± 164.7 µ , about 5 times the characteristic length of the micromotor. To achieve better performances, in 2018 they developed a model-based tracking system able to achieve ultra-fast localization of millimeter-scale hydrogel grippers from B-mode images with an average position tracking error of 25 ± 7 µ (over one percent the dimension of the gripper). 85 The same group recently demonstrated 3D position control of a magnetic sphere (800 µm in diameter) under US guidance. 86 To achieve 3D localization with 2D US tomographic images, the transducer was swept along the vertical direction to identify the horizontal plane containing the microrobot. The target plane (2D image) was thus processed by a tracker to extrapolate the remaining two coordinates. Promising results were reported also in ex vivo studies where different tasks of millimeter-scale robots were monitored through B-mode images. Qiu and coworkers exploited B-mode images for the localization of a helical propeller (characteristic length of 2 mm) moving in rat liver. 87 Cappelleri and coworkers used B-mode imaging to monitor a sub-millimetric tumbling microrobot capable of releasing the therapeutic payload in the dissected colon. 88 Sitti and coworkers used B-mode images to visualize the multimodal locomotion of a milli-robot within ex vivo chicken tissue. 14 However, none of these microrobot was smaller than 800 µm.
Recently, B-mode imaging has been successfully employed in microrobot swarm tracking. [89][90][91] Due to their enhanced area density, swarms of microrobots have shown very good contrast properties when imaged in B-mode. 92 In 2018, Zhang and coworkers reported the controlled navigation of a rotating colloidal swarm of paramagnetic NPs in a centimeter-scale arena, using B-mode images.
They found the dynamic equilibrium conditions of the swarm to produce periodic changes in their contrast. These changes can be used to discriminate the swarm from other elements in the image.
In addition, they demonstrated how the centripetal forces of the rotating swarms can be exploited to trap tunable concentrations of contrast agents (e.g. microbubbles) in their cores. 93 This feature can be exploited to further enhance the contrast properties of microrobotic swarms under B-mode imaging. In more recent studies, the same group exploited B-mode images to localize swarms of microrobots ex vivo in the bovine eyeball ( Figure 4C) and swine bladder. 94,95 On one hand, enhanced penetration depth could be achieved with passive US mode. With such a technique the microrobot itself acts as an US emitter by incorporating a micro-transducer (e.g. a cantilever) in its architecture. 98 The emitted signal can be tracked through the body by placing receiver antennas around the patient. 99,100 In principle, since the ultrasonic signals only need to travel through the media once, this approach offers twice the penetration depth than pulse-echo mode. However, it needs sufficient energy at the microrobot location to generate appropriate US emission.
Alternatively, ultrasound localization microscopy (ULM) is a promising approach to achieve superresolution. In ULM, the US signal scattered by microbubbles is processed along thousands of acquired frames to resolve images below the acoustic diffraction limit. With this technique, Lin and coworkers obtained US images with a pixel size of 10x10 µm using a 8 MHz transducer (featuring a diffraction limit of 100 µm). 101 Although promising, ULM has some bottlenecks for real-time MRs tracking. First of all, it is strongly dependent on a high dosage of contrast agent, not easily equipped on board a MR and secondly, it features poor temporal resolution. In fact, each ULM image was processed from 8000 acquired US frames, meaning more than a minute of acquisition with standard settings. Yu and coworkers 102 proposed a deconvolution method with which they reduced the acquisition time for each ULM image to about 6 s and other groups are working in the same direction, 103-105 but more efforts are required to make this method real-time.

Doppler Imaging
This US-based imaging technique relies on the Doppler effect, by which US waves that interact with moving objects experience a shift in their travelling frequency. More specifically, if two pulses are shot consecutively with a certain pulse repetition frequency (PRF) and interact with an object that has moved in the direction of wave propagation, the two echoes will show a difference in travelling frequency that is linearly related to the velocity of the moving object. This effect applies to all motion directions by considering their projection along the wave propagation direction. The Doppler signal is extracted through quadrature demodulation techniques 81 and is converted to a measure of the velocity of the imaged objects. The PRF, which relates to the temporal resolution is actually the Doppler signal sampling rate. According to the Shannon theorem such sampling rate sets the maximal detectable Doppler shift. In the past years, Doppler imaging has been used for the visualization and control in 3D of vibrating tethered robotic devices such as microneedles 106 and instruments for cardiac interventions. 107 More recently, the use of this technique in microrobotic applications has been investigated. In 2019, Sitti and coworkers proposed hairbot, a microscale device derived from hair tissues. 96 The functionalization of such devices with superparamagnetic iron oxide NPs (SPIONs) enabled the interaction with stem cells in microenvironments through magnetic actuation. Doppler imaging was used to image the hairbots after injection in ex vivo chicken breast ( Figure 4D). Other studies reported promising results in the fabrication of nanoagents for continuous Doppler contrast. Liberman and coworkers synthesized gas-filled silica nanoshells that can be detected in vivo using color Doppler imaging with great lifespans (over 11 days) ( Figure 4E). 97 These studies highlighted how, at least in principle, microrobots could be functionalized with Doppler contrast agents to achieve robust localization in tissues. Unlike Bmode imaging, the Doppler principle does not involve the amplitude of backscattered US waves, thus making it an effective tool for detecting moving objects in dynamic and echogenic environments, as body tissues are. The works related to US imaging (B-mode and Doppler) of MRs are summarized in Table 1.

Optical-based Techniques
Optical imaging utilizes non-ionizing radiation sources such as lasers or LEDs. Such techniques are relatively fast, safe and low cost and offer high spatial (order of few µm) and temporal (order of millisecond to seconds) resolutions. However, most optical techniques suffer from low penetration depth in biological tissues and are still limited to the investigations at sub-surface levels, typically between hundreds of micrometers and rarely beyond 1 mm deep. 108,109 Light penetration into tissues is limited by scattering and absorption at different biomolecules and endogenous chromophores and varies with tissue properties. These properties can be described by the mean free path (MFP) which is the average distance a photon travels between two scattering or absorption events and is defined as: where μs and μa are the scattering and absorption coefficients, respectively. 109 One strategy to overcome the limited penetration depth of optical techniques is to choose an appropriate wavelength window which can significantly reduce the tissue autofluorescence and light absorption. It is the case of the two biological windows in the NIR range respectively at 650-950 nm (NIR-I) 1000-1350 nm (NIR-II). For higher penetration depths, the use of wavelengths from the second biological window is desirable as it has been demonstrated elsewhere. 110 Hence, in the appropriate wavelength range, μs is much larger than μa for most of the tissues e.g. muscle, brain, breast and lung tissues, and the MFP can be approximated to 1/μs. Scattering in biological tissues occurs predominantly in the forward direction (direction in which the light strikes the tissue). This is considered by the anisotropy factor "g" which defines the degree of forward scattering (g=1 meaning the total forward scattering and g=-1 the total backscattering). Typical values for biological samples are in the range of g~0.9. The transport MFP (TMFP) considers the MFP and the average angle in which photons are scattered and is linked to the MFP by the reduced scattering coefficient μs′= μs(1-g). It is defined as TMFP = 1/μs′. 109 Another strategy to improve the penetration depth is to tune the emission features of target objects (i.e. MRs) by incorporating dedicated contrast agents or labels. For the sake of clarity in the classification, we have organized the optical imaging techniques used for MRs tracking into reflection-based and fluorescence-based techniques (see Figure 5A, i-ii).

Reflection-based Imaging
Optical techniques are widely used in body districts where the scattering of tissues is relatively low, e.g. in ophthalmology, sub-skin intervention and small animal experiments. As the penetration depth of light in tissue is comparably small, optical techniques are mostly used in reflection rather than transmission configuration, meaning that the illumination and detection are done with a singlesided optical access to the sample. RI does not rely on the emission of a fluorescent probe, but on the intrinsic reflection or backscattering properties of the sample. NPs or thin metal layers incorporated on target samples can be used to enhance signal reflectivity. In this regard, gold (Au) is one of the most suitable materials for its biocompatibility and high reflectivity over a broad spectral range. 111 However, when scattering scrambles both the illumination and the backscattered light, a high amount of background light occurs, which strongly reduces the signal to noise ratio (SNR). In the worst case, the scattered light saturates the detector and no sample information can be recorded. To reduce saturation and scattering, depth sectioning techniques can be employed. In this regard, confocal microscopy is based on filtering out-of-focus light using a pinhole. 112,113 Optical coherence tomography (OCT) is an interferometric technique that was developed by Fujimoto's group 114 in 1991 and takes advantage of the short coherence length of a light source which allows to record coherence gated images and obtain µm-scale cross-sectional imaging of the target. OCT provides improved spatial resolution (1-10 µm) in scattering tissue, but with high acquisition times due to the point-wise (time-domain OCT) or line-wise (frequency-domain OCT)scanning mode, thus avoiding real-time imaging. [115][116][117] Several studies reported the use of OCT imaging for tracking microagents in real-time. For instance, Li and co-workers employed a frequency-domain OCT setup to track magnetically-driven spherical microrobot (90 µm in diameter) in vivo in the portal vein of mice at a penetration depth of ~1.6 mm, with a line scanning rate of 5.5-70 kHz. 118 The same technique was also used to track reflective particles (polystyrene, 2 µm in diameter) as contrast agents to improve OCT imaging contrast in ex vivo mice liver and in vivo in zebrafish. 119 Likewise, SiO2 particles (1.2 µm in diameter, half-coated with a thin film of Au) were functionalized with a specific antibody that recognizes a biomarker (cardiac troponin I) of acute myocardial infarction to perform on-chip nonspectroscopic optical immunosensing. 120 By using the same principle, Nelson's group tracked the position of a microrobot (size: 285×1800 µm) for minimally invasive intraocular surgery. The magnetic robots were steered using rotating magnetic fields in ex vivo porcine eyes as well as in vivo in lapin eyes and consecutive images were acquired at a speed of 15 Hz. 48 The same group implemented optical microscopy combined with an electromagnetic coil system (8 coils arranged in a hemispherical configuration) and proposed a cylindrical millimetric (1x0.5 mm) robotic system for vascular retinal surgery in an eye model (human-like model eye) with and without eye's lens ( Figure 5B, i-ii). 121 An algorithm was proposed to localize the millimetric device based on its 3D structure. More recently, a similar concept was proposed to track magnetically driven micromotors employing IR light at a wavelength of 970 nm for performing in-situ microrheology analysis. 122 These reflective micromotors were successfully visualized with high spatial (about 20 µm) and temporal resolution (ms range) within scattering phantoms and ex vivo mouse skull tissues ( Figure   5C). The reported penetration limit was 0.32 TMFP, corresponding to about 160 µm depth in real mouse skull tissues. The main limiting factor of the employed setup was the limited dynamic range of the camera. Strategies such as pulsed laser excitation, and the usage of wavelengths of the NIR-II optical window would improve the achievable penetration depth. While the discussed methods suppress the scattered light, they result in a strong reduction of the SNR. In the Future Perspectives section, we will present methods that exploit the scattered light enabling increased SNR and penetration depth.

Fluorescence-based Imaging
FI is a widely used biomedical imaging modality; its working principle relies on the well-known Jablonski energy diagram 123  Standard fluorescent labels 127,128 have limitation issues associated with signal intensity strength, photobleaching, and short lifetimes. 129 Additionally, these fluorophores produce broad emission spectra that create overlapping detection ranges, making data analysis more challenging as highlighted by Nelson and coworkers. 42 To address these limitations, continuous development of species based on semiconductor nanocrystals have been carried out, in particular aimed at the development of a variety of quantum dots (QDs), 130   Although these techniques are well-established in the clinical practice, the employment of ionizing radiation-based techniques in the field of MRs tracking is still at its infancy and their application scenario should be carefully analyzed.

X-ray-CT Imaging
Radiography, fluoroscopy and computed tomography rely on the same electromagnetic radiations, namely X-rays (wavelengths in the range of 10-10 -3 nm). They differ essentially for the image acquisition timing, framerate and dimensionality, which is 2D in the first two cases and 3D in the latter.
The idea to use tomography in medical imaging dates back to 1895 when X-rays were discovered by W. C. Roentgen, but it took almost a century for the successful CT imaging device to be built. 134 X-rays are typically produced through a vacuum tube employing high voltage to accelerate electrons from a cathode to a tungsten alloy anode. During this acceleration process, the electrons release electromagnetic radiation in the form of X-rays. This kind of radiation transmits enough energy to ionize atoms and disrupt molecular bonds (also known as ionizing radiation). In general, such high energy allows high penetration depths, thus imaging of deep body regions. In X-rays imaging the sample is positioned between an emitter and a detector ( Figure 6A). An image of the sample is reconstructed through the differences in radiation absorption across different materials: materials with high absorption coefficients cause a reduction in the radiation dose reaching the detector and appear visible in the produced radiograph. Materials with higher density (ρ) or high atomic number (Z) such as Gold, Iodine and Barium, feature higher X-ray absorption coefficient μ, defined as: where A is the atomic mass and E is the X-ray energy.
Among the human tissues, bones are typically well visible through X-rays due to the presence of Calcium. When imaging microrobots with X-rays, we could expect higher contrast with respect to the surrounding tissues for metallic structures while soft, polymeric or hydrogel-based microstructures are less trivial to visualize. However, contrast agents typically employed in the clinics and base on high atomic number compounds (e.g. Iodine) significantly enhance image contrast when embedded within the microrobot structure.
As anticipated, radiography enables to acquire only static 2D images of selected human body regions. More interesting scenarios for MRs imaging and tracking are opened by CT and fluoroscopy. In the first case, an X-ray source and opposite detectors are rotated around a workspace while 2D scans are taken at discrete angular increments over a complete 360° rotation.
The projections are then processed to create a 3D reconstruction of the scanned object. The spatial resolution of CT imaging depends on the collimator dimensions, the number of detectors and the number of slices acquired over the 3D workspace. By decreasing the collimator size and increasing the number of detectors and 2D slices, a sub-millimetric resolution could be accomplished, accepting prolonged imaging time. 135 Overall CT is not suitable for real-time applications if high spatial resolution is required. Safety Although X-ray based imaging has been widely employed in the field of targeted therapy and microsurgery, e.g. for needle 137,138 or catheter steering 139  Overall, X-ray based techniques appear particularly interesting in terms of spatial and temporal resolution, potentially enabling real-time tracking and single microrobot/swarm imaging.
However, side effects associated with excessive exposure to X-rays should always be considered.
Nonetheless, it is worth underlining that, when properly balancing the exposure time, the X-ray energy and the frame rate (in the case of fluoroscopy), the delivered radiation dose can be acceptable with very low harm to body tissues.

PET-CT and SPECT Imaging
Nuclear imaging techniques rely on the detection of γ-rays (wavelengths lower than 0.1 nm). These imaging itself provides poor information on morphological and anatomical features and call for the employment of specific tracers (i.e. imaging agents) able to emit γ-rays. 143 PET and SPECT are minimally invasive high-resolution techniques which enable quantitative estimation of the number of radio-labeled species available in a specific location. Both techniques are widely employed in early diagnosis and in the evaluation of patient response to therapy by exploiting radiolabeled markers. 144 To these purposes, PET and SPECT are typically employed in conjunction with CT imaging to combine information on radioactivity distribution with anatomical ones.
The device for radiation detection is typically based on a scintillation camera which includes photomultiplier tubes to convert γ-ray photons into an electric signal. Radioactivity 3D distribution is assessed by combining projection images acquired through 360° rotations of the scintillation camera. The number and arrangement of employed detectors varies by technique, for instance assuming the sheet configuration in SPECT and the ring configuration in PET ( Figure 6A). In the case of SPECT, a lead collimator is also required.
Specific γ -emitting compounds enabling nuclear imaging are typically based on lighter elements produced via cyclotron, such as 11 C, 13 N, and 18 F in the case of PET and on heavier elements such as 99m Tc, and 123 I 145 with SPECT. PET radionuclides present very short half-lives ranging between ten minutes in 13 N and 110 minutes in 18 F if compared with SPECT radionuclides 99m Tc, and 123 I whose half-lives are 6 and 13 hours, respectively.
The γ-rays of PET generally possess higher energies and can travel through thicker tissues easily, experiencing lower attenuation compared to SPECT radiation. In addition, the production of two γ-ray beams in PET allows different detection modes (single incidence, true coincidences, etc.) leading to an increased sensitivity and spatial resolution. On the other hand, both positron emission and annihilation must occur to produce γ-rays in PET, generating the risk of position estimation artifacts caused by the positron travel between the two events. 146 Recently, some attempts to exploit nuclear imaging techniques to monitor and track MRs for medical applications have been reported. Sánchez and co-workers employed PET to track a large population of tubular Au/PEDOT/Pt micromotors (single tube length ~12 µm) in a tubular phantom. The absorption onto the micromotor surface of 124 I enabled PET tracking with 7 frames over 15 minutes. Results were confirmed by comparison with optical tracking, revealing the potential of such techniques towards in vivo microrobots tracking ( Figure 6C). 147 Iacovacci et al.  Table 1.
Overall, PET and SPECT offer great advantages with respect to other clinical imaging techniques especially in terms of penetration depth and selectivity (i.e. the ability to image specific elements).
On the other hand, these techniques have bottlenecks related to large acquisition time preventing real-time imaging, to the high cost of the imaging apparatus and to the limited life-time of radioemitting compounds. This makes these techniques more suitable for assessing an efficient target reaching and for post-operative monitoring than for fast position tracking. The radiation dose is also a critical point. However, when employing SPECT or PET for microrobots imaging, the absorbed radiation dose is significantly lower than the levels reached in X-ray CT. The reason is that in CT an external X-ray source/detector arrangement performs a 360° rotation to irradiate all the body's anatomy, while in nuclear techniques the source of radiation is limited to the very small volume of the microrobots.
In conclusion, PET potentially offers better spatiotemporal resolution and image quality in comparison to SPECT, but high costs and very short half-life of employed radioisotopes limit their inclusion in MRs for durable targeted interventions. Nonetheless, this aspect can make PET better in terms of safety for other interventions. In summary, PAI combines the advantages of light (high spatial resolution) with the penetration depth of US, 170,175 and spectral unmixing is a substantial feature to extract optical signatures of MRs from the surrounding tissues. It would be even advantageous to combine real-time US/PA with pre-acquired MR images towards more efficient diagnosis and therapy. 176 Another study highlights the dual-imaging system based on PAI and OCT for surgical guidance. 177 However, many applications require real-time imaging well beyond the penetration depths of PAI which is still challenging. Hybrid US and PA imaging (Figure 7D, ii) provides another interesting concept of functional and anatomical imaging, where US allows for accurate real-time tracking of MRs in deep tissues and PA is responsible for distinct signature discrimination from the surrounding tissues. In this regard, a submillimeter microrobot was imaged in opaque phantoms below 1.5 cm depth, with a spatial resolution of about 125 µm. 178 Recently, dual high-frequency US and PA imaging was used for the monitoring of swarms and single magnetically-driven spherical MRs (100 µm in diameter) in phantoms, ex vivo and in vivo (in mice bladder and uterus) intended for targeted drug delivery. 179 Furthermore, multi-wavelength excitation and spectral unmixing of injected MRs in tissue provided distinct spectrum signature compared to the tissue background after 3D reconstruction (Figure 7D, iii). The graph highlights the PA signal strength of the injected micromotors (yellow), oxy (red) and deoxygenated (blue) hemoglobin (Figure 7D, iv).  (Figure 8B).
In contrast to MPI, in MMUS the signal detected from the tracers motion does not allow to quantitatively estimate their distribution in the FoV. The reason is that NPs motion depends not only on magnetic quantities (e.g. magnetization curve, magnetic excitation signal) which are known, but also on the concentration of NPs that effectively accumulate in the site and on the viscoelastic properties of the surrounding medium which are usually unknown. Current research is aiming at developing MMUS algorithms for quantitative assessment of NPs distribution (known as the inverse MMUS problem) by comparison of MMUS images with offline simulated models ( Figure 8C). 187 Assessing the spatial resolution of MMUS is not straightforward. On one hand, the pixel size of a MMUS image is the same as in a standard B-mode image. On the other hand acoustic phase-shifts analysis enables to detected motions also in the nanometer range. 188,189 In this sense, this technique could be suitable also for nanoscale objects detection.
Temporal resolution is one of the bottlenecks of this technique. Currently available MMUS platforms require several US frames (called ensembles) to elaborate a MMUS image with acceptable SNR. Common algorithms running on standard platforms require an ensemble at least 5 s long to elaborate a clear MMUS image. 190 Different groups are currently investigating solutions for improving temporal resolution of MMUS imaging. [191][192][193] Overall, this technique provides the ability to detect motion events down to the nanometer scale.
Although currently used for characterizing tissues in quasi-static conditions, it can be potentially exploited for dynamic tracking of moving MRs, as for examples swarms in dynamic equilibrium.
However, more research efforts are needed to obtain better performances in terms of temporal resolution and to progress towards real-time monitoring. Future studies shall also address the robustness of this technique to physiological and biological motions in clinically relevant scenarios (e.g. due to breathing and hearth beating). 194 The works related to hybrid imaging (PAI and MMUS) of MRs are summarized in Table 1. The X-rays feature high resolution and deep penetration in body tissues but require expensive instrumentation. Furthermore, potential ionization can eventually cause damage to the medical personnel/patient, calling for the need of tuning radiation dose and exposure time. On the other hand, optical imaging methods based on fluorescence or reflection are somehow safe and carry high spatial resolution but have reduced tissue penetration which limits their application to subskin/superficial levels. MRI enables deep tissue imaging without any contrast agent along with acceptable spatial and temporal resolution but requires strong magnetic fields and expensive infrastructure. US alone suffers from spatial resolution but provide real-time monitoring and deep tissue penetration together with low cost and high safety. Table 1  While addressing future research in this field, it is worth considering that swarm control somehow smooths the technical requirements needed for MRs imaging. In real clinical settings and therapeutic paradigms, probably MRs swarm or larger carriers transporting multiple cells will be likely employed to effectively deliver the required therapy to the target region. Due to their increased spot size, swarms of MRs relax the requirements for high spatial resolution. In addition, swarms can potentially enhance the contrast between MRs and soft tissue and enable a more robust localization within the body. In fact, tissue intrinsic contrast and physiological motions (e.g. breathing, heart beating) pose an additional challenge in tracking moving MRs in vivo, which calls for the development of efficient contrast mechanisms to equip the MRs with. This considered, swarm control appears as a promising strategy due to enhanced contrast to some imaging techniques (e.g. US-based, like B-mode, Doppler, MMUS), and for the easiness of embedding higher doses of the contrast agent within the swarm formations.

FUTURE PERSPECTIVES
As highlighted in the previous sections, there is no "hero" imaging technique that can outperform all the others. Instead, cleverly designed hybrid imaging techniques and innovative labels are being explored to move the technological boarders towards exciting application areas. Such perspectives will be analyzed in the following paragraphs.

Contrast Enhancing Labels
One of the big difficulties and challenges for in vivo MRs applications is performing non-invasive,  Table 2. A multi-modal MR is also proposed embedding frontier contrast mechanism for multifunctional medical MRs (Figure 9).
MRI relies on contrast mechanisms to enlarge the difference in the relaxation times between different objects and its specificity may be improved through the development of relevant contrast agents. Current research is investigating the potentialities of MRI contrast mechanisms based on Quadrupole Relaxation Enhancement (QRE). Specifically, the interaction between water protons and suitable quadrupolar nuclei can lead to QRE of proton spins (provided the resonance condition between both spin transitions is fulfilled 195 ). This effect could be utilized as a frequency selective mechanism in T1 shortening contrast agents for MRI. Theoretically, the contrast enhancement factors of QRE are smaller than those of typical paramagnetic contrast agents. However, this mechanism depends on the characteristics of the applied external flux density, a property that can be exploited to enhance the relative contrast of properly designed materials with special fieldcycling MRI scanners. 196  Concerning MPI, suited contrast agents exhibiting very steep magnetization curves may be explored to further increase spatial resolution. It has been demonstrated that increasing the NP core size improves the magnetic susceptibility (steepness of magnetization curve) and leads to higher spatial resolution. However, this effect is limited by the so-called relaxation wall (25 nm core diameter), which causes blurring in images due to relaxation losses. 198 Recent studies highlighted that low-amplitude and low-frequency drive field parameters help mitigating relaxation losses. 199 A combination of larger NP core sizes and optimized drive field parameters may lead to the development of improved contrast agents, allowing to enhance MPI spatial resolution and make a considerable step towards MR tracking and control applications. High atomic number elements play a major role in ionizing radiation-based techniques.
Radioisotopes are fundamental imaging agents for PET and SPECT. The possibility to produce radiocompounds with specific chemical properties enables better inclusion of the imaging agents in the MRs structure and higher stability. Furthermore, as commonly done in the clinical practice, the possibility to functionalize such radiotracers to enable the binding with specific ligands so as to perform target-specific imaging, suggests a wide plateau of possibilities in the field of MR mediated targeted theranostics. [211][212][213] Elsewise, NPs made from high atomic number elements such as gold, bismuth, ytterbium and tantalum can be employed as x-ray contrast agents to improve the spatial resolution of these techniques. In fact, it has been demonstrated that such NPs provide better imaging properties, longer blood circulation times and lower toxicity than conventional iodinated X-ray contrast agents. 214 Additionally, the combination of two or more of these elements into a single carrier allows for the development of multimodal and hybrid contrast agents, for instance exhibiting optimal contrast properties over a range of different X-ray tube potentials or being suitable for more than one imaging modality (e.g. combination of X-ray and MRI). A good perspective is also represented by the potential translation of X-ray luminescent (XL) and X-ray fluorescence (XF) into biomedical imaging domain. These computed tomography techniques use external X-rays to stimulate secondary emissions (either light or secondary X-rays). These modalities surpass the limits of sensitivity in current X-ray imaging and promise to break through the spatial resolution limits of other in vivo molecular imaging modalities. 215 Smart materials can be investigated to explore the use of these phenomena in microrobotics applications. Reprinted in part with permission. 217

Recent trends in Hybrid Imaging
Modern imaging techniques that take advantage of different physical principles to gain in both spatial resolution and penetration depths are nowadays in the pre-clinical stage (e.g. PAI,). These hybrid techniques seem promising for testing MRs in vivo while controlling their motion and therapeutic tasks performance. Scientists from different disciplines are working on manipulating light, sound, and magnetic fields to overcome the diffraction and scattering limitations of current techniques. This can be possible by employing hybrid techniques or by functionalizing the MRs with absorbers, emitters or reflectors to enable improved sensitivity and contrast. In the following, innovative and promising hybrid techniques will be described to provide useful hints to the field of micro-and nanorobotics. Figure 10 shows basic principle and schematic illustration of such imaging concepts.

Acousto-Optic Imaging
Acousto-optic imaging (AOI) 221,222 does not require any optical absorption (differently from PAI) but is based on the acousto-optic effect. 223

Magneto-Optic Imaging
Magneto-optic imaging (MOI) is another hybrid imaging technique that combines an electromagnetic effect (eddy current or magnetic flux leakage) and the Faraday effect to generate a two-dimensional image. This technique was proposed and developed in the 1990s. 224 The propagation of light through a magnetic field is used to image target objects (e.g. defects or cracks in the surface of an object). A pulse generator sends an alternating current to the excitation coil which induces eddy currents in the sample. These eddy currents induce in turn magnetic fields when encountering any defect or damage in the sample surface. If the sample is illuminated by polarized light, the defect-induced magnetic field modulates the polarized light that now contains the information on the magnetic field distribution, which correlates to the surface defect. The reflected polarized light is captured by a CCD or CMOS camera to form a 2D real-time image of the target (Figure 10B).
Until now MOI has applications in detecting surface defects or cracks and corrosion in aircraft skins for testing purposes. 225 By using a pair of polarizers one can visualize and quantify the field distribution across the sample surface and by detecting the stray magnetic fields the magnetic properties of the materials can be detected. The MOI technique was mainly dedicated to identify defects/cracks in thin films. 226 By taking advantage of this, the MRs can be designed with small cracks/defects during the fabrication process using various lithography techniques (e.g. photolithography, 3D lithography, e-beam lithography or stamp printers). Of course, limitations imposed by light penetration have to be considered to verify the feasibility of this approach in a target district. Future study shall focus on imaging magnetic MRs and fabricating suitable designs of such robots thus sufficient reflected polarized light is detected from the target robot for MOI visualization. Additionally, image processing methods should also be improved to enhance MOI system for better sensitivity and image resolution.

Magneto-Acoustic Imaging
Magneto-acoustic tomography (MAI) employs high frequency (MHz range) pulsed magnetic fields to induce vibrations within tissues loaded with magnetic NPs. Particles vibrations produce mechanical waves which are registered through an external transducer and processed to reconstruct the magnetic material distribution (Figure 10C). Differently from MMUS, in which motions induced by continuous low frequency magnetic fields are detected through pulse-echo US sequencing. In MAI the transducer is only used as a receiver. This allows the MAI to have improved imaging bandwidth, limited mainly by the receiving bandwidth of the transducer, leading to better resolution. In addition, as compared to traditional US imaging, acoustic waves in MAI travel shorter paths and experience reduced attenuation in tissues, which leads to improved imaging depth. Because of the pulsed nature of the magnetic excitation signal, MAI allows also to reduce thermal issues related to continuous excitation, featured for instance by MMUS. In addition to basic MAI, magneto-acoustic tomography with magnetic induction has also been proposed (MAI-MI).
This latter method combines pulsed fields with static ones to generate acoustic waves through Lorentz force, which can be processed to estimate with high resolution the conductivity distribution of the source. Till now the technique has been employed to either estimate the distribution of magnetic NPs in the FoV, as is the case of simple MAI, 227 or for assessing tissue bioimpedance through MAI-MI. 228

Improvement of Optical Techniques with Wavefront Shaping
Coherent light that scatters during propagation through tissues forms a speckle pattern. The seemingly random speckle pattern is however deterministic as long as the scattering medium is static. [231][232][233][234][235] The key is the usage of spatial light modulators (SLMs) that are pixelated devices that enable to manipulate the phase of light. If an appropriate mask is displayed on the SLM, it is possible to remove the influence of a scattered field or to pre-shape the incident wavefront to achieve targeted delivery of a light-field distribution through thick scattering tissues ( Figure   10D). 231,[234][235][236][237] The huge technological and methodological progress of wavefront shaping techniques allows today for targeted light delivery deep into or through thick biological tissues. It was shown that even entirely scattered light can be controlled and used for imaging or focusing through thick scattering tissues. The only pre-condition is that the mask that is displayed on the SLM is appropriate and fulfills the desired task. It is by far the greatest challenge to obtain information on this phase mask and many solutions like optical phase conjugation, 236,[238][239][240] iterative optimization wavefront shaping, 241,242 and transmission matrix-based 243,244 approaches have been used to achieve this task. It was verified that light can be focused through approx. 100 TMFP, which corresponds to 9.6 cm tissue thickness. 245 However, the above mentioned approaches require physical access through the scattering medium, which makes most of them currently not suitable for real life applications. Most of the work reported in this field showed in a proof-of principle manner that the penetration depth of optical techniques could be dramatically increased using wavefront shaping. The correcting phase mask is just valid, as long the scattering medium is static. After a movement, the calibration has to be repeated. Hence, it is highly desirable to perform the calibration in-situ. The required refresh rate depends on the decorrelation time. It was shown in living mouse that the decorrelation time is between 50 ms and 2.5 s depending on the degree of immobilization and that there is also a strong dependence on the thickness and the type of the tissue, and the setup. 246 Tracking moving samples through thick scatterers is commonly achieved via speckle-correlation techniques, or differences of speckle fields, which yields the movement information, like velocity and moving direction. [247][248][249][250] However, with such techniques the exact position and movement in three dimensions is difficult to be extracted quantitatively. A possible way to achieve imaging and tracking in deep-tissue in a minimally invasive manner could be the usage of thin lens-less endoscopes, with diameters <350 µm, that are driven with wavefront shaping techniques. 251,252 Thus, wavefront shaping is a promising tool that will surely lead to further improvement in microscopy techniques, enabling imaging in deep tissues. Employing SLMs allows to do nearly anything with light waves, even to transfer image information through centimeters of tissue. The main task is to find strategies, how to get the required information to determine the appropriate phase mask. When wavefront shaping approaches will be optimized for tracking MRs, this will surely be a step towards clinical applications, as the fundamental limit of light penetration in tissue is not the physics but the technology and methodology in this still infant research topic.

AUTHOR INFORMATION
Spatial resolution: The minimal object size guaranteeing imaging or alternatively, which is the minimal distance possible between two objects to be discriminated from each other.

Temporal resolution:
The minimal time required to complete the data acquisition needed to reconstruct one complete image of the object.

Penetration depth:
The measure of how deep the probing signal of an imaging technique can penetrate through tissues. It can be defined as the distance where the primary intensity is reduced to l/e (i.e. 37%) of the initial transmitted intensity.

Field of view:
The dimensions of the exact anatomic region included in an image.

Contrast agents:
A substance to enhance the contrast of structures, bodily fluids or body organs using medical imaging techniques.
Hybrid imaging: It refers to the fusion of two or more imaging techniques to give birth to a more versatile imaging modality.