UWB Bowtie Antenna for Medical Microwave Imaging Applications

This work presents a design and experimental validation of a compact ultra-wide band (UWB) bowtie antenna and balanced-to-unbalanced circuit (balun) for medical microwave imaging applications working in the frequency band 1-6 GHz. The UWB balun is perpendicularly attached to the planar bowtie arms, whose dimensions were miniaturized by rounding the bowtie edges. The antenna reflection coefficient was lower than -12 dB for tissues with a higher water content. The antenna possesses a high radiation efficiency (over 80%) and very low backward radiation. The predicted and measured SAR distributions together with the electric field |E| distribution proved a symmetrical radiation pattern. The potential of the presented antenna element was demonstrated by successful reconstruction of the images from the measured data by microwave imaging methods. For this purpose, a microwave imaging system equipped with eight antennas was implemented. The radar approach was used to determine the region of interest (ROI) where the area for microwave tomography reconstruction was identified. The dielectric properties within the ROI were reconstructed by using Born approximation method at 1 GHz. This confirmed that the antenna can be used as the basis for more accurate multi-frequency MWI systems and can be implemented in UWB MWI hybrid systems.


I. INTRODUCTION
ICROWAVE imaging (MWI) methods in medicine have been explored intensively in last two decades as an alternative to current conventional imaging methods such as Magnetic Resonance Imaging (MRI), Computed Tomography (CT), Positron Emission Tomography (PET), etc. [1]. The advantage of MWI methods is its relatively short acquisition time and operating costs [2]. Furthermore MWI utilizes nonionizing radiation and its systems can be implemented in a small and compact way [3]. MWI have the potential to be effectively applied directly in ambulance vehicles in the diagnosis of brain stroke, resulting in a significant reduction in the time between the onset of the stroke and its treatment. This is the key aspect to eliminating negative health consequences and to obtaining a better patient prognosis. Another possible application of MWI systems is non-invasive temperature monitoring during hyperthermia treatment in which the temperature of the target tumor region is artificially increased to the range of 40-44°C for at least one hour [4]. This temperature increase can be detected as a function of changes in dielectric properties using the MWI technique. The MWI methods are based on the exposure of the imaging domain with electromagnetic (EM) waves and the measurement of reflected and scattered electromagnetic waves at interfaces with different dielectric properties usually using antenna arrays. The region of interest can be scanned and visualized in 3-D through various reconstruction algorithms (depending on the microwave imaging method) processing information about EM waves received by individual antenna elements. Such 3-D reconstructions are reported in [5] or [6]. The image reconstruction capability of the MWI system is greatly influenced by the characteristics of the antenna elements we used. The selection of an appropriate antenna element for MWI system depends on the selected mode of microwave imaging. First, the microwave tomography (MT) method uses signal processing in the frequency domain and applies methods capable of estimating the dielectric profile of the scanned region. The antennas of MT systems are usually placed equidistantly around the imaging domain [5]- [10]. Second, radar-based (confocal) imaging methods process signals in the time domain and can locate dielectric inhomogeneity (scatterers) within a tissue, without the ability to reconstruct the dielectric profile. Radar-based reconstruction algorithms are usually significantly less demanding on computation and time. The basic principle is to transmit ultra-wide band (UWB) signals, providing high imaging resolution. The transmitted signal can be formed by short periodically repeated pulses, pseudorandom binary sequences, etc. Such radar-based systems for biomedical imaging are presented in [11]- [13]. By combining these two principles, i.e. the microwave tomography and radar-based approach, it brings the possibility to reduce the analyzed tissue area to the less voluminous region containing significant inhomogeneity. Reduction of the imaged area can effectively speed up the MWI reconstruction [14], [15]. These novel combined MWI systems lead to new requirements for the antenna elements.
Among UWB antennas that are suitable for microwave imaging at microwave frequencies are Vivaldi planar antennas [15], [16], double ridged horn antennas [17], [18], planar monopoles [6], [19], [20] and bowtie dipoles [21], [22]. The Planar Vivaldi antenna proposed by Burqui et al. [16] uses exponentially tapered antipodal arms to radiate energy directly to the tissue. This antenna was designed for the breast cancer detection system based on the radar approach. Due to the very low relative permittivity of the breast (approx. εr= 6), canola oil is used as a matching liquid. UWB ridged horn antennas have high radiation efficiency and the backward and side radiation is suppressed. These ridged horn antennas are usually filled with high permittivity material (e.g. ceramics [18]) inserted inside the antenna and reducing the antenna dimensions. The weight of the antenna is high and due to its filling is also difficult to manufacture compared to other eligible antennas for microwave imaging. Planar UWB monopole antennas, for example, from Jafaris et al. [19] offer a low-cost solution of the problem, as they are easy to fabricate. Planar monopoles do not need a matching liquid because they can be placed on the surface of the analyzed tissue. However, the advantage of the simplicity of planar design entails the drawback of power supply in the plane of the monopole patch, which is not very appropriate and makes the use of such a monopole in an array difficult from the array perspective. Furthermore, the radiation pattern of the planar UWB monopoles varies quite significantly with frequency [20], which may cause distortion in the time domain. Bowtie dipoles, generally planar dipoles and its modifications are simple to fabricate. The geometry is broadband and the antenna is fully attachable directly to the tissue (the matching liquid is not needed). Compared to the horn and Vivaldi antennas, the bowtie antenna has a lower radiation effectivity and is more sensitive to the surrounding environment. Due to the high permittivity value of the muscle tissue, a higher part of the electromagnetic energy is emitted towards the phantom, which improves the efficiency. However, the bowtie antenna requires symmetrical feeding, which is usually suitable only for a limited frequency range. Table 1 compares UWB antennas for microwave imaging devices with here mentioned UWB antenna.
The purpose of this work was to design an UWB bowtie antenna including UWB balun for combined microwave tomography. In the first step we have studied and compared several possible microstrip UWB baluns (Section II. A and B). The most promising balun was combined with the bowtie antenna, whose dimensions were minimized by adjusting the shape of bowtie edges using a numerical parametric study. In the next step, we numerically tested basic antenna parameters for antenna usability. We numerically tested antenna radiation efficiency, verified that bowties arms are symmetrically fed by visualization of surface current density and |E| field distribution. The sensitivity of the antenna to air gaps formation and as well as the signal distortion of the UWB signal were tested (II. A, Part 3). We compared predicted and measured reflection characteristics and SAR distribution of manufactured UWB bowtie antenna at the phantom of different equivalent tissues (II. A, Part 4). Then we equidistantly positioned eight manufactured UWB bowtie antennas into the MWI head system. The potential of this system to detect and reconstruct the dielectric inhomogeneity inside the human head phantom using a combination of radar and MT approaches was studied (Section III.).

II. ANTENNA DESIGN
The antenna presented in this paper is intended for the biomedical microwave imaging applications. Microwave imaging systems are mostly composed of an array of antennas that are attached to the area to be imaged. The antenna parameters influence the reconstruction/imaging ability of the entire system. The antenna used in UWB MWI system should satisfy multiple demands in the intended ultra-wide frequency band, which are: • Efficient transmission of EM energy into the tissue -low reflection coefficient of the antenna, minimized radiation to the surrounding regions. • Adequate near field radiation patterns -wide beam width in the azimuth plane and acceptably narrow beam width in the elevation plane. • Minimal (time domain) distortion of the transmitted/received signals and short impulse response to achieve high spatial resolution of the system. • Acceptable physical dimensions of the antenna element.
The dimensions of the antenna determine the maximum number of antennas that could be distributed around the area to be imaged. According to our experience, the minimal number of antennas for successful MWI of i.e. head with decent accuracy in 2D is 8 (in single ring arrangement). We estimate that up to 16 bowtie antennas can be positioned to cover the entire head area (in 3D).

A. Methods -Antenna Design
In the first step of the UWB bowtie antenna development, we designed the bowtie arms shape. Then we created UWB microstrip balun, as a bowtie dipole is a symmetrical structure, and requires balanced feeding. In the third step, we optimized the entire UWB bowtie antenna geometry to fulfil given performance requirements. The commercial EM numerical simulator COMSOL Multiphysics (COMSOL AB, Sweden) based on the Finite Element Method (FEM) and Sim4life (Speag, Switzerland) based on the Finite Difference Time Domain Method (FDTD) were used for the antenna design and parameters optimization. In all numerical simulations, the antenna was directly attached to the homogeneous phantom with dielectric parameters being similar to the specific human tissues (see Fig. 1). The Fig. 1 shows used relative permittivity and electrical conductivity of considered tissues in the whole simulated band, data were adopted from [29].

1) Bowtie Antenna Arms
A bowtie antenna is a planar structure usually designed as a one-layer PCB. We have considered several eligible shapes of bowtie arms available in the literature [30]. We tested following geometries, namely "rounded bowtie antenna" (REBA), "triangle bowtie antenna with rounded corners" (TBARC) and fundamental "triangle bowtie antenna" (TBA), see Fig. 2. All shapes were derived from the TBA geometry with total length and width of 30 mm. Bowtie arms were placed at 1.524 mm height substrate with dielectric properties corresponding to the Rogers RO4003C (Rogers Corporation, USA). In the simulated antenna the angle  was fixed to 45° as a whole antenna structure together with balun showed the best reflection coefficient. The bowtie antenna performance is also dependent on the dipole rounding radius R. The parametric study has been carried out to optimize the impedance matching of the proposed antenna over frequency band of interest.

2) UWB Balun
UWB balun (balanced to unbalanced) symmetrization circuit is a key feature in the design of UWB antenna with symmetrical radiating pattern. The most applicable UWB balun solutions are ferrite or planar tapered baluns. We decided to use the planar tapered UWB balun over ferrite baluns. This decision was based on assumption that the ferrites baluns can transform only limited RF power. According to our experience, the planar tapered UWB baluns also cause lower UWB signal distortion at higher frequencies in comparison with the ferrite baluns. We calculated the strip widths of microstrip line and doublesided parallel strip line (DSPSL) in a way to obtain 50 Ω impedance at used 0.508 mm height Rogers RO4003C substrate. The impedance characteristic was calculated according to the formulas presented in [31]. The balun length was set to 48 mm as a quarter of wavelength at 1 GHz. The width of the microstrip ground plane was set to wg = 25 mm. Four different balun shapes were considered and the corresponding numerical models were created. The baluns were modelled as mirrored pairs with microstrip line at both ends and the double-sided parallel strip line in the middle (in total 2 x 48 mm long). Geometries of the baluns are shown in Fig. 3. For all balun versions S-parameters were calculated in the frequency band of 0.2 -6 GHz. The antenna for the medical microwave imaging should be directly attachable to the imaged area. If this option is to be maintained, then it is convenient to feed the bowtie dipole in a direction perpendicular to the plane of the dipole arms. The balun was attached perpendicularly to the bowtie dipole arms (TBARC with a radius of 6 mm).

3) Simulation of Antenna Properties
The following experiments were designed to clarify UWB antenna parameters, such as antenna usability for MWI of different tissues, antenna sensitivity to lower phantom contact to tissue, antenna radiation efficiency and antenna radiation symmetry. These experiments were performed in the environment of COMSOL Multiphysics (FEM-based) as numerical simulations and in the Sim4life numerical simulator (FDTD based). First, a simple numerical model consisting of a UWB antenna surrounded by a sphere of a radius of 20 cm where the front hemisphere was defined as the brain phantom. To acquire further insight into antenna performance, we have studied the current distribution over the surface of the bowtie arms. The current distribution was visualized using vectors to prove the ability of the balun to provide sufficient symmetric feeding of the bowtie arms. Second, we evaluated the radiation efficiency of the antenna. The introduced antenna is intended to operate in a lossy medium (average human brain phantom) in the near-field antenna region. To express the antenna radiation efficiency to the human brain phantom, the parameter Ef was introduced. The parameter Ef was calculated according to equation (1) where Pinput (W) is the input power, Phead (W) is the power absorbed in the brain phantom in front of the antenna. The method of radiation patterns, with respect to antenna theory [32], [33], is (for this case) hardly applicable for the estimation of radiation efficiency. Instead, we proposed to evaluate Specific Absorption Rate -SAR (W/kg) in yz plane for frequency range 1 -6 GHz with a 2 GHz step. The numerically obtained SAR was normalized by the maximum value and compared with the measured SAR. The boundary conditions of the model were set to scattering boundary conditions that suppress reflections. We further numerically evaluated the normalized magnitude of the intensity of the electric field intensity |E| depending on the radiation angle (°) in xz and yz planes (H-and E-plane). The electric field intensity was computed for the frequencies 1, 3 and 5 GHz in both planes perpendicular to the antenna ( Fig. 4 (a) -red curve is for E-plane and green curve for H-plane). The evaluation angle ranged from -90° to 90 ° with respect to the antenna position. The transition between brain and air hemisphere stands for 0° angle. Furthermore, the antenna medical versatility was verified numerically and experimentally via the comparison of the reflection parameters following the setup in the Fig. 5. In this setup the whole antenna, including the SMA model, UWB balun, and dipole arms, is faced with a single tissue cubic phantom. The numerical analysis is performed in the Sim4Life software as the phantom mimics the brain, bone, skin, and muscle dielectric properties (relative permittivity and electrical conductivity -presented in Fig. 1) [29]. The antenna ability to transmit and receive an ultra-wide band signals with low distortion was tested using the same simulation setup described above (see Fig. 5). The UWB signal that is emitted and which is propagating through lossy media (i.e. human tissues) is attenuated. This attenuation is frequency dependent and affects the signal amplitude and signal shape. To eliminate this effect, and only for this signal distortion study, the phantom electrical conductivity was set to 0 (S/m). Thus, we can observe just signal distortion caused by antenna element itself. One antenna was determined as transmitting (Tx) one. It emits the UWB Gaussian pulse of the frequency band 1-6 GHz. The opposite antenna was in the receiving mode (Rx). This analysis was done using the Sim4Life software, ZMT Zurich Med Tech AG, Switzerland. The most common parameter for UWB signal distortion evaluation is fidelity, which is calculated  z x y x y z as a cross-correlation between the normalized Tx and Rx signals. The fidelity factor F (-) is defined as follows [36]: where T(t)is transmitted signal, R(t) is received signal, t is time and τ is time displacement. The time domain signals from proposed antenna are also compared to signals from previously used UWB bowtie antenna (for brain imaging) as well as the fidelity factor F. Subsequently, two additional numerical studies of the antenna sensitivity to the lower phantom contact (partial air gaps at the antenna edges formations which can occur in real applications) 1) on flat ( Fig. 6 (a)) and 2) rounded phantom ( Fig. 6 (b)) were performed. The dielectric parameters of the phantom were set as human brain (see the dielectric parameters in Fig. 1). Firstly, we tested four cases of reduced contact between antenna and flat brain phantom. The ratio of contact with phantom and air for different cases is listed in the following Table 2 and is visualized in the Fig. 6 (a) -case A. In all simulated cases the part of the phantom under the antenna was cut off and replaced by air (situation A-D). We observed the influence of the partial contact on |S11| parameter of the antenna. In the second experiment, the phantom was attached to the series of rounded brain phantoms of different diameter (see Fig. 6 (b)). Generally, we tested six phantoms of diameters from 100 -350 mm with 50 mm step. This should represent a real clinical situation when the antenna will be attached to the rounded part of the body (i.e. head, arm, leg etc.).

4) Measurement of Antenna Properties
The SAR in the brain phantom was measured by certified cSAR3D (Schmidt & Partner Engineering AG, Switzerland) [34], which is containing the flat head phantom (Head Tissue Simulating Medium -HSL) setup together with vector network analyzer (VNA) ZNB8-32 (Rohde & Schwarz, Germany) used as a microwave signal generator, see diagram of setup in the Fig. 4 (b). The dielectric properties of the HSL phantom are as human head ±10 % [34]. The dielectric properties are identical as presented in Fig. 1 for brain tissue. The output power of the VNA was set to 13 dBm. The antenna was directly attached to the center of the cSAR3D phantom. Good adhesion of the antenna to the phantom was achieved by applying a very thin water layer to avoid air gaps creation. For the measurement of the antenna |S11| parameter, the fabricated dipole antenna was placed directly in contact with the phantom. The head phantom preparation is further described in Section III. A. The agar muscle phantom was prepared according to the recipe presented in [35]. The VNA R&S FSH8 (Rohde & Schwarz, Germany) was used for reflection coefficient measurements in the band 0.5 GHz -6 GHz.

B. Simulation Results -Antenna Design 1) Simulation of Bowtie Antenna Arms
The results of the simulation of mentioned geometries in the frequency band of 0.5 -6 GHz are presented in Fig. 7 (a) and (b). Obtained | 11 | curves showed that TBA shapes are better impedance matched at lower frequencies, but all simulated shapes have sufficient reflection coefficient (| 11 | lower than -10 dB). Due to the demands on the compact antenna size in the imaging system, the TBARC geometry with a rounding radius of 6 mm was chosen for the proposed antenna, as this geometry enables the reduction of the bowtie arms width from 30 mm in the TBA geometry to only 13 mm and offers the lowest predicted reflection coefficient from all studied shapes.

2) Simulation of UWB Balun
S-parameters (modules of S11 and S21) of all four tested baluns were numerically calculated and are presented in the Fig.  8. Best results were obtained with the exponential tapered balun, which was adopted from [16]. The balun geometry was modified to achieve better reflection coefficient within used   Fig. 3). The influence of small variations of the microstrip ground plane width wg on balun performance were found to be insignificant. The optimized and finalized balun geometry with dimensions is presented in the Fig. 9. The exponential balun offers the highest and frequency stable transmission coefficient together with the lowest reflection coefficient.

3) Finalized Antenna Dimensions
The finalized geometry of the designed UWB bowtie antenna with inserted balun between the bowtie arms is optimized for operation in the frequency bandwidth of 1-6 GHz. The resulted antenna model is shown in the Fig. 10. The antenna is of compact size (60x60x50 mm) with the possibility of further dimension reduction, as the conductive bowtie arms motive was minimized as described earlier.

4) Simulation of Finalized Antenna Properties
The surface current density J (A/m) distribution in the indicated frequency bandwidth on the bowtie arms of UWB antenna is shown in Fig. 11. The module of surface current density |J| is represented by color map. The white cones are representing the normalized vectors of surface current density J at instant when current density shows a maximum. The white line is representing the 10 % isocontour of the maximum value of the surface current density. To maintain the whole frequency band, the results at frequencies 1 -6 GHz are shown in the Figs. 11 (a-f). The maximum current surface density is distributed along the edges of the antenna in proximity to the antenna excitation. From the Fig. 11 (a-f) it can be concluded that for all frequencies the vectors and modules of the surface current density are symmetrical for both arms.   Fig. 8. The simulated module of (a) reflection coefficient (|S11|) and of (b) transmission coefficient (|S21|) of baluns of different ground transition shapes (geometries are presented in Fig. 2).
x (mm) x (mm) In the Table 3 the ratio of absorbed power in the brain phantom to input power -parameter Ef at frequencies 1 -6 GHz is presented. The data in the table express how much of the input power is effectively transferred towards the head phantom. The absorbed power consists of power being reflected in the antenna structure due to the impedance mismatch, from power being radiated outside the phantom (backward radiation) and dissipated in the antenna structure. Based on described power calculations, proposed UWB antenna shows ability to deliver above 80 % of the input power to the analyzed phantom. The developed UWB antenna delivers on average 20 % more energy within the whole frequency bandwidth than our previously used bowtie antenna presented in [37]. Instead of the antenna radiation pattern analysis in the far field, we evaluate the antenna radiation based on comparison of normalized magnitudes of the electric field in lossy brain phantom along the antenna radiation angle for different frequencies.
The resulted E-field shapes are presented in the Fig. 12 (a) for Eplane and (b) for H-plane. Resulted |S11| parameters for antenna being attached to the specific human tissues as brain -blue curve, bone -orange curve, skin -violet curve and muscle -green curve + waterred curve are shown in the Fig. 13. According to the results listed in the Fig. 13, the antenna, attached to the brain and muscle phantom, is usable to work within the frequency band 0.   [37] and black curve represents signal received by here presented UWB antenna. Substituting to the equation 2, the calculated fidelity factor F in the brain phantom for the previously used bowtie antenna is 0.86 and for new proposed UWB antenna is 0.81. The fidelity factor of novel antenna is slightly lower than previously used antenna -higher fidelity factor means lower signal distortion. The fidelity factor of the new proposed antenna in case of other tissues as skin and muscle are 0.79 and 0.83 respectively. For bone tissue the fidelity factor was 0.70. This is corresponding with results of |S11| parameter when the antenna was attached to these tissues (Fig. 14). However, in brain, skin and muscle the new antenna shows the ability to transmit and receive the UWB signal without any significant distortion effects and with low subsequent ringing. The Fig. 15 shows values of |S11| parameter when the antenna was attached to the flat phantom with reduced contact from 100 % to 50 %. For contact 100 % (blue curve), 95 % (red curve), 75 % (yellow curve) the antenna is working in the whole frequency band without any significant changes. For lower contacts 55 % and 50 %, the |S11| parameter is under -10 dB within the frequency band 0.5-3.1 GHz. For antenna contact lower than 55 %, the matching liquid is required. The results in Fig. 16 show the |S11| parameter of the antenna attached to the rounded phantom of different diameters in the range of 100 -350 mm. A smaller phantom diameter results in a lower contact between the antenna and the phantom. But for all cases the central active part with metallic bowties was attached properly. For phantoms of diameter 100 -300 mm the change in the phantom diameter has no significant influence on the |S11| parameter. For higher diameters (350 mm and more), we can expect reduction of the usable frequency band, and therefore the matching liquid will be required. This effect can be caused by the formation of surface waves that propagate at the air-phantom interface.

C. Antenna Fabrication
The antenna design is composed of several planar parts, which were fabricated at 1.542 mm and 0.508 mm (UWB tapered balun) high substrate Rogers RO4003C using the standard PCB technology realized by Pragoboard ltd., Czech Republic. The    substrate Isola DURAVER (height 2 mm) was used for the mechanical reinforcement of the balun. The mechanical connection was made by standard two component epoxy glue.
To the feeding point the SMA connector was soldered. Photo of fabricated UWB bowtie antenna is shown in the Fig. 18.

D. Antenna Measurements Result
The simulation results presented in the Fig. 13 were confirmed by a measurement of |S11| parameter on brain and muscle phantom. The real measurements performed using liquid (brain phantom -blue dashed line) and solid phantom (muscle phantom -green dashed line) are indicated by the dashed line.   Fig. 17 shows predicted (a, c, e) and measured (b, d, f) SAR distributions under proposed UWB antenna (YZ plane) in the human brain phantom for frequencies 2, 4, and 6 GHz. The SAR distributions show good homogeneity and symmetry at all discussed frequencies compared to the numerical simulation. The differences between the simulated and measured SAR distributions are mainly visible in the close proximity of the antenna.

III. UWB MICROWAVE HEAD IMAGING SYSTEM
The potential of the proposed UWB bowtie antenna was tested in the microwave tomography system developed previously in our laboratory.

A. Methods -MW Brain Imaging System
The MWI system was previously described in [7] including Gauss Newton MWI reconstruction technique. The system has been modified for an array of eight UWB antennas. This was done with respect to cross-coupling reduction among antennas. The number of used antennas was reduced from 10 to 8 to increase the angle between antennas and thus to reduce the antenna crosstalk. The dimensions of the container are 200 x 160 mm based on average adult male head [38]. The antennas were equidistantly placed in one ring in the plastic antenna holder to form a multistatic array around the area to be imaged. The whole imaging system is shown in the Fig. 19 and 20. The container height is 200 mm height with 2 mm thick wall. It consists the octagonal pins at the bottom, which are used for precise phantoms positioning which represents inhomogeneity to be imaged. These cylindrical phantoms have diameter of 40 mm. All plastic parts of containers were printed using the PETG material at Prusa i3 MK3 (Prusa Research, Czech Republic) 3D printer. The antennas were inserted and attached to the wall of container by plastic screws. The mentioned VNA ZNB8-32 connected to the microwave switching matrix ZN-Z84-B42 (Rohde & Schwarz, Germany) was used to measure the transmissions between antennas (S-parameter matrix). UWB bowtie antennas were connected to the switching matrix using a semi-rigid coaxial cable.

1) Liquid Phantoms
A liquid phantoms were used for all measurements since they enable to attach the antenna without any air gaps and allow us to place desired inhomogeneity to a predefined position. We prepared phantom with the dielectric parameters of average human brain according to the dielectric properties taken from [29]. The phantom was prepared by mixing of deionized water and isopropyl alcohol. By changing the isopropyl alcohol and deionized water ratio an additional phantom was prepared representing small deviation in dielectric parameters (change in phantom relative permittivity εr was approximately + 5 %). The dielectric parameters of mixed liquid phantom samples were measured using commercial system SPEAG DAK-12(Schmidt & Partner Engineering AG, Switzerland) at the frequency range 0.2 -3 GHz just before measurements by the MW imaging system. The dielectric probe was connected via coaxial cable to the handheld VNA Keysight N9913A (Fieldfox, USA). Measured data were analyzed using SPEAG software to calculate the dielectric properties. The final phantom composition is listed in the Table 4. The Cole-Cole model was fitted to that measured data by using Levenberg-Marquart algorithm, where the initial values were set from [29]. The whole procedure is further described in [39]. Dielectric parameters were extrapolated to the frequency band of our interest (0.5-6 GHz).    Fig. 21 (a) representing relative permittivity and Fig. 21 (b) electrical conductivity of the prepared liquid phantoms. Blue curve corresponds to the phantom of human head, the red curve is presenting the phantom with approximately 5 % change in dielectric parameters. This modified phantom was used as an inhomogeneity inserted into the head phantom during simulation and measurement sessions of the imaging system. Each curve is representing the mean value calculated from 10 measurements, while the shading error bars (within the measured frequency band 0.2-3 GHz) are representing the ± combined standard uncertainty type C. The composition of each phantom components in weight percent is listed in the Table 4.

2) Measurement procedure -MWI System
The container of MWI system was filled with described liquid brain phantom (Fig. 20 (b)). The plastic cylinder filled with inhomogeneity phantom, representing the inhomogeneity was inserted in predefined position in container. The whole system is shown in the Fig. 19 and 20. Four positions of this cylinder were measured to demonstrate the ability of the MWI system to reconstruct 2D scattering phantom profile, respectively the ability to detect inhomogeneity within the analyzed region. The following measurement parameters were set: frequency bandwidth 1-6 GHz, VNA output power 13 dBm, intermediate frequency bandwidth 100 Hz. Following our previous experiences, we neglect frequencies above 6 GHz, which are significantly attenuated in the lossy materials as human brain and thus they do not provide useful information for image reconstruction. All antennas were used in transmit/received mode in which we systematically changed the transmitting single antenna while all others antennas were set to receive mode. We measured two scenarios, 1) container filled fully with brain phantom and 2) container filled with brain phantom with inserted inhomogeneity phantom at desired place. This ensured that all scattered and reflected signals were caused just by the inhomogeneity phantom in the cylinderclutter and background signals were removed. The plastic container for inhomogeneity phantom was present during all measurements. During the measurement 1) the inhomogeneity container was filled by brain phantom and during the measurement 2) the inhomogeneity container was filled by inhomogeneity phantom.

3) Signal Processing and Image Reconstruction
• Radar Approach The resulting frequency domain data were transformed into the time domain using Inverse Discrete Fourier Transform (IDFT). Thus the system impulse response matrix was obtained. Time-shift ( ) for each captured signal and each focal point r0 was calculated by estimating propagation speed vp of the electromagnetic wave in measured tissue with dielectric constant εr and from known round-trip distance dm of transmitting antenna, of particular investigated focal point and receiving antenna [2]: where c (m/s) is speed of light and is a relative permitivity of phantom at central frequency of measured bandwidth. The "Delay and Sum" algorithm was used for the inhomogeneity position reconstruction. By applying this procedure, 2D energy profile showing scattering/reflecting areas was created. The intensity I(r0) in particular focal point r0 was obtained through the following formula (4): where Tw is a predefined time-window and wm is a weighting factor for path dependent attenuation, t is time, N is number of channels and ∆ is a differential signal calculated as: where si is a signal measured when the inhomogeneity phantom was present, sw is a signal measured without inhomogeneity phantom. The signals captured on the antennas with an internal angle wider than 90° to the transmitting antenna have been omitted as they usually do not improve the final image quality [40].
• MT Approach To demonstrate MT approach, we implemented differential microwave tomography algorithm presented by Scapaticci et. al. [41] and used by Tesarik et. al. [42] for non-invasive microwave thermometry. This algorithm exploits Born approximation (BA) with regularization by Truncated Singular Value Decomposition (TSVD) to reconstruct differential dielectric profile of brain phantom with inhomogeneity placed inside. According to traditional Volume Integral Equation (VIE) defined e.g. in [43] a total electric field of a multiviewmultistatic system is equal to the sum of an initial electric field generated by antennae of the system and scattered electric field from objects placed inside the system. The initial electromagnetic field was computed by solution of forward problem using COMSOL Multiphysics (FEM -Finite Element Method). The model in COMSOL is presented in the Fig. 20 (a). Then the VIE can be modified [43], [44] to linear equation (6) ∆ = • ∆ , where ∆ is the differential S-matrix calculated as difference between measured S-matrix without any inhomogeneity inside the system and S-matrix with inhomogeneity inside the system. Le is a so-called linear operator built from computed electric fields by the forward solver and ∆ (∆ , ∆σ) is unknown differential object function. TSVD solves equation (6) where number of pixels of ∆ is usually larger than the number of independent values of ∆ .

• Hybrid Approach
The combination of radar and MT approach can optimize the process of image reconstruction in terms of speed and accuracy. We can split the process into two steps. First, from the results obtained from the radar-based approach, the algorithm identifies the ROI (region of interest) where the inhomogeneity inside the system is placed. Second, this a priori information can be used as input for forward solver or algorithm during MT image reconstruction process, respectively. Complex permittivity reconstruction using differential MT approach with BA and TSVD regularization is sensitive on artefacts in the resulting image. Usually one or more false reconstructed "hot spots" can be observed. Therefore, the real position of inhomogeneity could be difficult to identify and also lower number of truncation level needs to be applied. It results to distortion of reconstructed values of differential dielectric parameters.
Using the information about ROI obtained from radar-based approach the MT reconstruction process is made only in that ROI which allows to use higher values of truncation number and thus improves the reconstructed values of differential relative permittivity as well as conductivity. The reconstruction algorithm is also speed up because of using computed initial electric fields only from ROI and thus reducing the size of Le as well as ∆ . Considering the iterative MT reconstruction algorithms as Gauss-Newton with Tikhonov regularization presented for instance in [7], the hybrid approach offers rapid time reduction of estimation of dielectric parameters. To find the values of complex permittivity inside the 8-port MT system using 8 iterations took approx. 20 hours. Commonly in each interaction the whole forward numerical model must be solved. If model is limited to exact ROI defined by radar approach the solution time as well as image reconstruction by MT approach is shortened.

B. Reconstruction Results -MW Brain Imaging System 1) Radar Approach
Four different inhomogeneity phantom positions (P1-P4) were measured as it is shown in the Fig. 19. The 2D reconstructed images of four scattering profiles, shown in Fig. 22, were obtained by differential signals processing of the measured impulses (radar method). The position of the inhomogeneity phantom within the container is in each case marked by the red circle. The reconstructed images are presented for clarity by isocontours. The results are showing that by radar method we are able to detect the position of the phantom with 5 % change (approx. + 2.5 in εr) in dielectric parameters. For all four positions the maximum intensity is detected within the red circle which is indicating the actual phantom position. It took 90 seconds to measure and 22 seconds to reconstruct one inhomogeneity position. For two inhomogeneity positions P1 and P2 we detect our ROI marked by pink rectangle in the Fig.  22. These ROI were determined as double distance between the isocontour 0.9 and 0.5 to all directions.

2) Hybrid Approach
The measured representative |Snn| and |Smn| parameters for case when the inhomogeneity was placed in position P4 (according to Fig. 19) are presented in the Fig. 23. The blue curve is representing the |S33| parameter (in that case, antenna 3 is the nearest antenna to the inhomogeneity position P4). The blue shading errorbar is representing the standard deviation of all |Snn| parameters -of all antennas in the system (A1-A8). The representative transmission coefficients are presented by red (|S32|), orange (|S34|) and violet (|S37|) curve. The lowest transmission coefficients are identical for all here presented signals (between 3.5 -5 GHz), where the transmission coefficients reach values around -90 dB. At the frequency around 4 GHz the |S37| exceeds the value -100 dB. Absolute value of differential transmission coefficients |ΔSmn| between the cases with inhomogeneity (inhomogeneity position P4) and without inhomogeneity are in the Fig. 24. Based on ROI's defined by radar approach around positions of inhomogeneity P1 and P2, we reconstructed the differential relative permittivity   as well as electrical conductivity profiles inside the ROI's using MT approach. We used the operation frequency of 1 GHz as reasonable frequency regarding the space resolution and penetration depth of EM wave. The truncation level of TSVD was set to 28 where the maximal value is equal to 36. The reconstructed images are depicted in Fig. 25. We can see that both positions of inhomogeneity were correctly identified by MT algorithm based on differential relative permittivity data. The reconstructed contrast of differential relative permittivity approx. + 2 especially in P1 ROI was in very good match with measured value + 1.56 by commercial dielectric probe. In the image also the artefact of negative differential contrast -2 was reconstructed and is visible in the right down corner of P1 ROI. Due to radar-based ROI identification this artefact was eliminated, and true position and contrast of inhomogeneity was determined. The reconstructed differential electrical conductivity data was distorted and neither the data limitation by ROI helped to get satisfying results.

IV. DISCUSSION
The UWB bowtie antenna for application in UWB MWI systems was designed and its calculated characteristics confirmed by series of numerical simulations and measurements. Bowtie arms were miniaturized by edge rounding which is beneficial in MWI systems with higher number of antennas. Smaller dipole arms dimensions have a positive influence on the antenna impulse response. We designed the UWB balun which provides sufficient reflection coefficient (|S11| < -10 dB) in its entire frequency band (1-6 GHz) when connected to the bowtie dipole arms. Here presented UWB antenna stands out above the other UWB antennas designed for microwave imaging [23]- [28] especially in the wide usable frequency bandwidth (1-6 GHz) which provides sufficient resolution for most biomedical application of microwave imaging. The comparable antennas intended for microwave imaging are listed in the Table 1 with their basic parameters. To prove this statement the final antenna element was attached to the various type of tissues and the |S11| parameter was simulated. The simulation results show that the antenna is fully functional (|S11| < -10 dB) in its entire frequency band (1-6 GHz) for tissues that have higher water content (brain, muscle). For skin tissue the frequency band decreased slightly to 1-5 GHz. For tissues (bone tissue in our case) with low water content, we believe that the frequency band is still useful between 1 and 4 GHz. The predicted UWB bowtie antenna |S11| characteristic was confirmed by measurement by using muscle and brain phantom (within the imaging system) in the 1 -6 GHz frequency band with magnitude well below -12 dB in the whole studied frequency band. The fractional bandwidth of our antenna is 143 %. The fractional bandwidth (and subsequently applicable frequency bandwidth) is one of the highest values which can be found in the literature. This ensures short impulse response of the antenna. Only the antenna presented in [28] is offering higher fractional bandwidth (163 %) with comparable fidelity factor and is designed for the head imaging. Our UWB antenna possesses a high radiation efficiency (over 80 % in the whole frequency band) to the brain phantom and therefore does not suffer from high backward radiation. The antennas with high effectivity are able to transmit higher energy and thus detect lower dielectric changes. Simulated and measured SAR distributions as well as |E| field distribution in the brain phantom of proposed antenna shown symmetrical radiation pattern. Final antenna dimensions are 60x60x50 mm. The front panel of dimensions 60x60 mm was chosen as appropriate for insertion to the brain imaging system previously developed in our lab. However, the simulation of the antenna sensitivity to air gaps between antenna and brain phantom showed that in presence of air under active part (metallic bowties) the antenna performance is low and therefore application of matching liquid is required. If the air gap is outside the antenna active area, that air gap has no or very low effect on |S11|.The fidelity factor for brain phantom was 0.81, for skin 0.79 and for muscle 0.83. In case of bone tissue, the fidelity was lower (0.70). According to our experience, the fidelity factor around 0.80 is perfectly sufficient for MWI techniques and offers adequate spatial resolution. Due to the large UWB bandwidth the fidelity factor is lower compared to the antennas previously published (fidelity factor in papers [23]- [28] ranges from 80 % to 92 %). This ensures that presented antenna has minimal time domain distortion of the Tx/Rx signals. Developed UWB bowtie antenna was implemented in eight channel microwave UWB imaging system. Results showed very good detectability of the inhomogeneity position within head phantom by the radar approach. The ROI was selected using reconstructed data from the radar approach. This ROI was used as an input a priori information to MT approach based on Born approximation. This led to an increased accuracy of the reconstruction as well as filtering of unwanted hotspots. The reconstruction process was also speeded up by approx. 20 % due to reducing the investigation domain only on ROI. The reconstruction of measured data based on electrical conductivity profile was distorted and not fully successful. It could be caused by too high truncation level. Since the conductivity changes were in orders of tenths, the stability of the algorithm could be oversaturated. Reduction of truncation level would lead to decrease of reconstructed contrast accuracy of relative permittivity data. Increasing the antennas number and its positions in different rings can lead to better reconstruction accuracy. The scattering profile reconstruction shows that also for very low change in dielectric parameters enabling the system to accurately detect the stroke position.

V. CONCLUSION
In this work, the design of an UWB antenna for MWI was presented and its suitability was experimentally verified. The presented UWB antenna shows a high radiation efficiency (above 80%) and a symmetrical radiation pattern in the whole working frequency band. The useful frequency band is very wide (1-6 GHz) which allows the antenna to be used for most microwave imaging applications offering high spatial resolution. The fidelity factor over 80 % ensures good antenna functionality with low distortion and good impulse response. The antenna parameters such as good |S11|, working frequency band and radiation efficiency are important for the sensitivity of the reconstruction algorithms. Our antenna has fulfilled all parameters mentioned above in the whole UWB band. The compactness of the antenna predestines it for MWI systems with a higher number of antenna elements, and the construction of the antenna then enables cheap and repeatable production by established methods of production of printed circuit boards. The antenna can thus be used as the basis for more accurate MWI multifrequency systems and can also be implemented in UWB MWI hybrid systems combining microwave tomography and radar approaches.