**5.3. Breast and body surface reconstruction**

### *5.3.1. Method*

302 Ultra-Wideband Radio Technologies for Communications, Localization and Sensor Applications

possibility of mechanical scanning.

**5.2. Breast and body phantoms** 

the imaginary part

phantom surface (insert).

However, we prefer this measurement scenario for our current investigations, and intend to weaken the contact problem in the future by 2 or 3 different array sizes and an additional gentle suction of the breast into the antenna array by a slight underpressure. In section 5.4, we present an experimental measuring set-up where we pursue a strategy of nearly direct contact imaging in order to conjoin the advantages of contact-based imaging with the

In the context of UWB tissue sensing, the water content plays a key role as it determines the inherent dielectric properties (´ and ´´) [43]. Moreover, the water content is known to vary among the different human tissues as well as between specific normal and pathologic ones, thus offering a potentially broad spectrum of UWB applications for biomedical diagnostics. Oil-in-gelatin phantoms, mimicking the dielectric properties of human tissues, were manufactured according to a protocol from [24]. The water concentration varied between 19 and 95% (v/v; ~ 10% water graduation steps), to obtain a set of materials with different permittivity values (´ ranging from 8 to 59 and ´´ ranging from 0.5 to 11, both averaged over frequencies from 1 to 4 GHz). The measurements were carried out by means of the M-sequence devices [76], [87] with HaLoS chipsets and a frequency bandwidth of 4.5 GHz, as well as the radar data acquisition and analysis software "ultraANALYSER" developed for this purpose.

The variation of the oil-water-concentration led to the identification of distinct permittivity values ´ (Fig. 44, insert) of the different oil-in-gelatin phantoms. The phantom, which was manufactured without oil (95% water), showed values between 53 and 59 for the real part ε´ and between 11 and 10 for the imaginary part ´´ of the permittivity in the frequency range between 1 and 3.5 GHz (Fig. 44, insert). The results for pure distilled water are also

**Figure 44.** Dielectric properties of nine oil-in-gelatin phantoms with varying percentage of water (from

Error bars represent the standard deviation from an average of three individual measurements on the

of the phantoms. Both parts increase with an increasing water-concentration.

and

19% to 95% water (v/v)) and porcine muscle tissue. Depicted is the correlation of the real part

displayed. The real part of permittivity agrees well with literature data [88].

The benefits of the exact knowledge of the breast surface for non-contact microwave breast imaging are manifold and can improve the results significantly. The inclusion of the breast shape information is essential to calculate the wave traveling path in order to image the interior of the breast based on radar beam-forming techniques. Some approaches use the surface information for initial estimations. Other non-contact measurement approaches strive to illuminate the breast from a specific distance which requires a very fast online surface identification in order to adapt the antenna position during measurement. Furthermore, in the case of varying distances between antenna and breast, the exact knowledge of the breast surface can improve the estimation of the skin reflection component for a better early time artifact removal. In order to reduce the calculation time, the region of interest (i.e. the region for which the image has to be processed) can be restricted based on known surface geometry [89], [90].

Additionally to the significance for breast imaging, UWB microwave radar is suitable for whole body surface reconstruction which can be used in other medical microwave applications as well as in safety-relevant tasks, e.g. under-dress weapon detection.

The Boundary Scattering Transform (BST) represents a powerful approach for surface detection problems. BST and its inverse transform (IBST) were introduced 2004 by Sakamoto and Sato [91] as basic algorithms for high-speed ultra wideband imaging, called SEABED (Shape Estimation Algorithm based on BST and Extraction of Directly scattered waves). Since then, this idea has been extended from mono-static 2D-imaging to the point of bi-static 3D-imaging (IBBST) [92]. The SEABED algorithm represents a high–speed, high-resolution microwave imaging procedure. It does not include the entire radar signal; it uses only wave fronts instead. Furthermore, changes (derivatives) of the propagation time (transmitter object surface receiver) depending on the antenna position during the scan process play an important role. SEABED consists of three steps: 1. Detection of the wave fronts and calculation of their derivatives with respect to the coordinates of the scan plane. 2. Inverse Boundary Scattering Transform, which yields spatially distributed points representing the surface of the object. 3. Reconstruction of the surface based on these points.

The practical applicability of the original algorithm to the identification of complex shaped surfaces is limited because of the inherent planar scanning scheme and, therefore, the disadvantage of illuminating only one side of the object. For this reason, we extended the bistatic approach of [92] toward non-planar scanning and a fully three-dimensional antenna movement based on the idea that in the case of arbitrary non-planar scan schemes the current scan plane can be approximated by the tangential plane at each antenna position [93]. An antenna position dependent coordinate transform which ensures that the antenna axis is parallel to the *x* -axis and the current scan plane is parallel to one plane of the coordinate system allows the application of the IBBST for nearly arbitrary scan surfaces. More precisely, this generalized approach is limited to scenarios where the antennas will be

### 304 Ultra-Wideband Radio Technologies for Communications, Localization and Sensor Applications

moved orthogonally or parallel to the antenna axis, which is fulfilled in most practical cases. First results of breast shape identification were published in [94], [95].

Based on the following transform equation, the coordinate of the specular point can be calculated

$$\begin{aligned} \overline{X} &= \overline{X} - \frac{2D^3 D\_{\overline{X}}}{D^2 - d^2 + \sqrt{(D^2 - d^2)^2 + 4d^2 D^2 D\_{\overline{X}}^2}} \\ \overline{y} &= \overline{Y} + \frac{D\_{\overline{Y}}}{D^3} \left( d^2 (\overline{x} - \overline{X})^2 - D^4 \right) \\ \overline{Z} &= \overline{Z} + \sqrt{D^2 - d^2 - (\overline{y} - \overline{Y})^2 - \frac{(D^2 - d^2)(\overline{x} - \overline{X})^2}{D^2}} \end{aligned} \tag{5}$$

ultraMEDIS – Ultra-Wideband Sensing in Medicine 305

The main challenge is the exact detection of the wave fronts and their proper derivative. For the purpose of wave front detection, we use an iterative correlation-based detection algorithm similar to [96]. In this connection, a short antenna impulse response over a sufficiently wide angular range plays an important role. The difficulties of obtaining appropriate wave front derivatives result from the three-dimensional nature of the problem. The antennas are moved and the transmitted waves are reflected in the three-dimensional space. Especially in the case of wave front crossing and impulse overlapping as well as sparsely detected wave fronts, it is very complicated to recognize which identified wave front at one scan position is related to which wave front at the previous scan position and vice versa. So, it may happen that derivative values are wrongly calculated, which can lead to a spatially false projection of the surface points. In order to avoid such errors, we establish

thresholds of feasible derivative values dependent on the antenna beam width.

*General limit values:* The range of values of the distance derivatives *DX*,*<sup>Y</sup>* is theoretically bounded between 0 and 1 depending on the slope of the reflection plane (tangent plane of the object surface at the specular point). In the case of parallelism between reflection plane and antenna axis, 0 *DX* , whereas in the case of orthogonality, 1 *DX* . Thus, calculated values 1 *DX* are definitely caused by incorrect wave front detection. Consideration of these general boundaries and exclusion of wave fronts exceeding them yields a significant

*Customized plausibility limit values:* The boundary 1 *DX* assumes an antenna radiation angle of 90° or more, which is not given using directive radiators, e.g. horn antennas. In that case, the range of plausible derivative values can further be restricted. Assuming a

min sin *<sup>d</sup> <sup>D</sup>*

Wave fronts with lower *D* values would imply specular points which are located outside the

<sup>2</sup> <sup>2</sup> <sup>2</sup> <sup>2</sup>

*L x L x dd L x L x dd*

cos cos cos 2 cos cos cos 2 2 

with the perpendicular from the reflection plane to the distant antenna 2 2

its perpendicular angle *β* and the reflection angle *γ* as depicted in Fig. 45. This value yields

and a distance between transmitter and receiving

(6)

 

*L*

cos

, *d* and *D* can be

(7)

sin

,

*D d*

*D d*

antenna of 2*d* the minimum reasonable value *D*min can easily be defined by

Furthermore, a maximum distance derivative *DX* depending on

*5.3.2. Detection and elimination of improper wavefronts* 

improvement.

established:

max

maximum antenna radiation angle

antenna beam and, therefore, can be ignored [98].

*<sup>X</sup>*

 

*<sup>D</sup> <sup>x</sup>*

where *xyz* , , are the coordinates of the reflective surface point (specular point), *XYZ* , , are the coordinates of the center between the two antennas, *D* is the half distance transmitter reflection point receiver, *d* is the half distance between the two antennas, and *X dD <sup>D</sup> dX* , *<sup>Y</sup> dD <sup>D</sup> dY* symbolizes the derivatives of the distance with respect to the denoted direction of antenna movement. The bars above the symbols mark the coordinates of the transformed coordinate system [93].

**Figure 45.** Ray geometry of the inverse bi-static boundary scattering transform (IBBST)

The main challenge is the exact detection of the wave fronts and their proper derivative. For the purpose of wave front detection, we use an iterative correlation-based detection algorithm similar to [96]. In this connection, a short antenna impulse response over a sufficiently wide angular range plays an important role. The difficulties of obtaining appropriate wave front derivatives result from the three-dimensional nature of the problem. The antennas are moved and the transmitted waves are reflected in the three-dimensional space. Especially in the case of wave front crossing and impulse overlapping as well as sparsely detected wave fronts, it is very complicated to recognize which identified wave front at one scan position is related to which wave front at the previous scan position and vice versa. So, it may happen that derivative values are wrongly calculated, which can lead to a spatially false projection of the surface points. In order to avoid such errors, we establish thresholds of feasible derivative values dependent on the antenna beam width.

### *5.3.2. Detection and elimination of improper wavefronts*

304 Ultra-Wideband Radio Technologies for Communications, Localization and Sensor Applications

First results of breast shape identification were published in [94], [95].

3

*<sup>D</sup> y Y dx X D <sup>D</sup>*

*Y*

*x X*

calculated

*X dD <sup>D</sup> dX* , *<sup>Y</sup>*

*dD <sup>D</sup>*

transformed coordinate system [93].

moved orthogonally or parallel to the antenna axis, which is fulfilled in most practical cases.

Based on the following transform equation, the coordinate of the specular point can be

3 2 2 2 22 2 2 2

*D D*

*D d D d dDD*

( )4

*X*

22 2

*X*

(5)

2

( )( ) ( )

*dY* symbolizes the derivatives of the distance with respect to the denoted

( )

22 2

2 24

*D d xX z Z D d yY <sup>D</sup>*

where *xyz* , , are the coordinates of the reflective surface point (specular point), *XYZ* , , are the coordinates of the center between the two antennas, *D* is the half distance transmitter reflection point receiver, *d* is the half distance between the two antennas, and

direction of antenna movement. The bars above the symbols mark the coordinates of the

**Figure 45.** Ray geometry of the inverse bi-static boundary scattering transform (IBBST)

2

*General limit values:* The range of values of the distance derivatives *DX*,*<sup>Y</sup>* is theoretically bounded between 0 and 1 depending on the slope of the reflection plane (tangent plane of the object surface at the specular point). In the case of parallelism between reflection plane and antenna axis, 0 *DX* , whereas in the case of orthogonality, 1 *DX* . Thus, calculated values 1 *DX* are definitely caused by incorrect wave front detection. Consideration of these general boundaries and exclusion of wave fronts exceeding them yields a significant improvement.

*Customized plausibility limit values:* The boundary 1 *DX* assumes an antenna radiation angle of 90° or more, which is not given using directive radiators, e.g. horn antennas. In that case, the range of plausible derivative values can further be restricted. Assuming a maximum antenna radiation angle and a distance between transmitter and receiving antenna of 2*d* the minimum reasonable value *D*min can easily be defined by

$$D\_{\min} = \frac{d}{\sin \alpha} \tag{6}$$

Wave fronts with lower *D* values would imply specular points which are located outside the antenna beam and, therefore, can be ignored [98].

Furthermore, a maximum distance derivative *DX* depending on , *d* and *D* can be established:

$$D\_{\mathfrak{T}\_{\text{max}}} = \frac{\sqrt{\left(L + \cos \beta \cdot \omega x\right)^2 - \left(L + \cos \beta \cdot \omega x\right)\cos \beta \cdot 2d + d^2} - \sqrt{\left(L - \cos \beta \cdot \omega x\right)^2 - \left(L - \cos \beta \cdot \omega x\right)\cos \beta \cdot 2d + d^2}}{2 \cdot \omega x} \tag{7}$$

with the perpendicular from the reflection plane to the distant antenna 2 2 cos sin *D d L D d* , its perpendicular angle *β* and the reflection angle *γ* as depicted in Fig. 45. This value yields

max *DX* sin for mono-static arrangements ( 0) *d* and approaches to this value in the case of *L d* , respectively. For further details of the derivation of these thresholds and reconstruction examples illustrating the accuracy enhancement due to the application of these thresholds, we refer to [98].

### *5.3.3. Reconstruction results*

For repeatable measurements, we applied a female dressmaker torso which is filled with tissue-equivalent phantom material (Fig. 46). Based on linear and rotational scanners which can move or rotate the object and/or the antennas, several non-planar scan schemes can be realized in order to scan this torso efficiently. In the following, the results of breast shape identification based on a toroidal scan will be shown. The M-sequence radar device used has a bandwidth of 12 GHz [97].

**Figure 46.** Female torso filled with human tissue mimicking phantom material and delineation of the toroidal scan scheme to reconstruct the chest surface

ultraMEDIS – Ultra-Wideband Sensing in Medicine 307

an even more precise wave front identification. Naturally, this is only possible if the radar

Figure 47 shows the UWB reconstruction results of the mentioned torso in comparison to a laser reference measurement. In order to quantify the accuracy, the distances between each calculated UWB surface point and the laser-based detected surface is calculated. The resulting mean aberration lower than 1.4 mm underlines the potential of this method. Nevertheless, it is obvious that a further enhancement of the wave front detection represents a residual challenge in order to fill in increasingly the areas of sparsely distributed surface

**Figure 47.** Exact UWB chest surface reconstruction (black) and appraisal of performance values by means of a laser reference measurement (gray) showing a mean aberration lower than 1.4 mm.

Furthermore, the applicability of a 3D-IBBST-based UWB surface reconstruction method for medical applications other than breast imaging as well as for security scenarios (under dress

The main parts of UWB time domain imaging are the removal of clutter (also referred to as early time artifact removal) and beam-forming (also referred to as migration or back projection). Because the tumor reflections are overlapped by antenna cross-talk and skin reflection, clutter removal is a very important and critical component of signal

device fulfills such high time stability requirements.

weapon detection) is demonstrated in [98].

*5.4.1. UWB breast imaging in time domain* 

**5.4. Contact based breast imaging** 

points.

Numerical problems may arise in the calculation of derivatives from discrete data (discrete time intervals; discrete antenna positions in the space) which have to be considered for setting measurement and processing parameters. The resolutions of spatial scanning and radar signal sampling have to be harmonized carefully with each other in order to avoid derivative artifacts. The maximum possible error of the derivative is <sup>0</sup> <sup>ˆ</sup> <sup>2</sup> *<sup>x</sup> t v e D <sup>x</sup>* where

*t* is the time resolution of the wave front detection, *x* is the antenna displacement applied for the calculation of *Dx* and <sup>0</sup> *v* is the propagation velocity of the electromagnetic wave. Hence, it will be obvious to meet the requirement of for example ˆ 0.05 *<sup>x</sup> e D* (0.05 is more than 5 percent relative error with respect to *Dx* max for antenna beam widths < 90°!) with an antenna displacement such as *x* 2.5 cm in air 0 0 ( ) *v c* the wave front detection has to be realized with a time accuracy of 8.33 ps which has to be provided by interpolation within the wave front detection algorithm. Higher performance requirements presuppose an even more precise wave front identification. Naturally, this is only possible if the radar device fulfills such high time stability requirements.

Figure 47 shows the UWB reconstruction results of the mentioned torso in comparison to a laser reference measurement. In order to quantify the accuracy, the distances between each calculated UWB surface point and the laser-based detected surface is calculated. The resulting mean aberration lower than 1.4 mm underlines the potential of this method. Nevertheless, it is obvious that a further enhancement of the wave front detection represents a residual challenge in order to fill in increasingly the areas of sparsely distributed surface points.

**Figure 47.** Exact UWB chest surface reconstruction (black) and appraisal of performance values by means of a laser reference measurement (gray) showing a mean aberration lower than 1.4 mm.

Furthermore, the applicability of a 3D-IBBST-based UWB surface reconstruction method for medical applications other than breast imaging as well as for security scenarios (under dress weapon detection) is demonstrated in [98].

### **5.4. Contact based breast imaging**

306 Ultra-Wideband Radio Technologies for Communications, Localization and Sensor Applications

for mono-static arrangements ( 0) *d* and approaches to this value in the case

 *<sup>x</sup> t v e D <sup>x</sup>* where

of *L d* , respectively. For further details of the derivation of these thresholds and reconstruction examples illustrating the accuracy enhancement due to the application of

For repeatable measurements, we applied a female dressmaker torso which is filled with tissue-equivalent phantom material (Fig. 46). Based on linear and rotational scanners which can move or rotate the object and/or the antennas, several non-planar scan schemes can be realized in order to scan this torso efficiently. In the following, the results of breast shape identification based on a toroidal scan will be shown. The M-sequence radar device used has

**Figure 46.** Female torso filled with human tissue mimicking phantom material and delineation of the

Numerical problems may arise in the calculation of derivatives from discrete data (discrete time intervals; discrete antenna positions in the space) which have to be considered for setting measurement and processing parameters. The resolutions of spatial scanning and radar signal sampling have to be harmonized carefully with each other in order to avoid

*t* is the time resolution of the wave front detection, *x* is the antenna displacement applied for the calculation of *Dx* and <sup>0</sup> *v* is the propagation velocity of the electromagnetic wave. Hence, it will be obvious to meet the requirement of for example ˆ 0.05 *<sup>x</sup> e D* (0.05 is more than 5 percent relative error with respect to *Dx* max for antenna beam widths < 90°!) with an antenna displacement such as *x* 2.5 cm in air 0 0 ( ) *v c* the wave front detection has to be realized with a time accuracy of 8.33 ps which has to be provided by interpolation within the wave front detection algorithm. Higher performance requirements presuppose

derivative artifacts. The maximum possible error of the derivative is <sup>0</sup> <sup>ˆ</sup> <sup>2</sup>

max *DX* sin

these thresholds, we refer to [98].

*5.3.3. Reconstruction results* 

a bandwidth of 12 GHz [97].

toroidal scan scheme to reconstruct the chest surface

### *5.4.1. UWB breast imaging in time domain*

The main parts of UWB time domain imaging are the removal of clutter (also referred to as early time artifact removal) and beam-forming (also referred to as migration or back projection). Because the tumor reflections are overlapped by antenna cross-talk and skin reflection, clutter removal is a very important and critical component of signal preprocessing before beam-forming can be carried out. Most clutter removal approaches assume that the clutter appears very similar in each channel and, thus, its estimation improves with increasing channel number. It must be noted that this holds only for channels with comparable clutter parameters. That means clutter estimation and removal has to be done separately for groups consisting of only associated signals (channels with identical antenna distances and boresight angles Tx-Rx), which accomplishes this task. In scientific work on simulation, this circumstance is commonly ignored. For practical applications, however, it has to be taken into consideration.

ultraMEDIS – Ultra-Wideband Sensing in Medicine 309

' 50 , a

tissue properties and image processing. Here, we pursue the objective of very small antenna dimensions, short impulses and an application in direct or quasi direct contact mode.

Initially, we used short bow-ties (Fig. 48) with the dimensions of 8 mm x 3 mm implemented on Rogers® 4003 substrate (0.5 mm) using PCB technology. Dipoles have to be fed

These antennas cannot be matched over a large bandwidth, which leads to unwanted reflections between antenna and amplifier. There are two options concerning the handling of this problem: realization of a sufficient line length between antenna and amplifier (in order to gate out the reflections) or implementation of the amplifier circuits directly at the antenna feed point. On an interim basis, we pursued the first strategy using long cables

70 cm cable will ensure that any reflections from inside of the breast (diameter ~ 10 cm) and

As mentioned above, the contact between antennas and breast skin represents a crucial aspect for sufficient signal quality. Regarding clinical requirements (e.g. disinfection) we plan to place the antennas behind a thin examination mold. But this additional interface reduces the signal quality significantly. Therefore, a thin (~2 mm) matching layer consisting of tissue mimicking phantom material was inserted between the examination mold and the antennas in order to increase the signal energy penetrating the tissue and reduce the backward radiation (Fig. 49). The benefit achieved when using a thin contact layer was also

We built up two preliminary array set-ups for phantom measurements, both including eight antennas and distributing them around a circular segment (diameter 9.5 cm) in steps of 22.5°. An array with a horizontal antenna arrangement is shown in Fig. 50. Exemplary phantom measurement results achieved with these prototypes are published in [104] and

differentially. The balanced feeding is realized by differential amplifier circuits [103].

between antenna and amplifier. Assuming a maximum mean tissue permittivity

Therefore, we investigated the usability of small interfacial dipoles.

**Figure 48.** Small bow-ties on Rogers substrate

unwanted reflections at the amplifier do not overlap.

investigated and verified by simulations (Fig. 5 in section 2.3.2).

[105] and will be summarized in section 5.4.3.

The simplest approach is to estimate the clutter by means of the average value. Tumor reflections are assumed to appear uncorrelated in the channels and to be negligible in the averaged signal. Even though publications about advanced clutter removal algorithms emphasize the weak points of this self-evident approach, it must be noted that it works relatively robustly in the case of covering tumor response by clutter when some of the proposed alternatives are not applicable.

Image formation algorithms using time domain beam-forming can be included in the following generalized formula:

$$I\left(\mathbf{r}\_{0}\right) = \sum\_{\varepsilon\_{h}=-\overline{\varepsilon}\_{h}f2}^{\overline{\varepsilon}\_{h}f2} h\left(\mathbf{r}\_{h}, \mathbf{r}\_{0}\right) \cdot \left(\sum\_{n=1}^{N} \sum\_{\varepsilon\_{w}=-\overline{\varepsilon}\_{w}f2}^{\overline{\varepsilon}\_{w}f2} w\_{n}\left(\mathbf{r}\_{w}, \mathbf{r}\_{0}\right) \cdot S\_{n}\left(t + \tau\_{n}\left(\mathbf{r}\_{0}\right) + \tau\_{w} + \tau\_{h}\right)\right)^{2} \tag{8}$$

where N is the number of channels, *<sup>n</sup> S t* is the clutter subtracted signal of channel *n* , <sup>0</sup> **r** symbolizes the coordinates of the focal point (image position vector), <sup>0</sup> ( ) *<sup>n</sup>* **r** is the time delay of channel *n* related to the focal point at 0**r** and *I***r**<sup>0</sup> is the back scattered energy which has to be mapped over the region of interest inside the breast. Based on two FIR filters, the different extensions of the common delay-and-sum beam former can be expressed. Pathdependent dispersion and attenuation [99], [100] can be equalized by means of <sup>0</sup> , *wn w* **r** which – in the simplest case - can be only a weight coefficient. Other improvements can also be included by convolution in the time domain, e.g. the cross-correlated back projection algorithm [101]. *h <sup>h</sup>* ,**r**<sup>0</sup> represents a smoothing window at the energy level or a scalar weight coefficient [102].

### *5.4.2. Measurement setup based on small antennas*

The efficient penetration of the electromagnetic waves into the tissue and the spatial highresolution registration of the reflected signals are crucial tasks of the antenna array design. In this regard, efficiency is not only a matter of radiation efficiency or antenna return loss, respectively. An efficient antenna array design concerning biomedical UWB imaging purposes comprises also the shape and duration of signal impulses, angle dependence of the impulse characteristics (fidelity), and the physical dimensions of the antenna. These interacting parameters are hardly to accommodate to each other within one antenna design. Generally, compromise solutions have to be found considering basic conditions of scanning, tissue properties and image processing. Here, we pursue the objective of very small antenna dimensions, short impulses and an application in direct or quasi direct contact mode. Therefore, we investigated the usability of small interfacial dipoles.

**Figure 48.** Small bow-ties on Rogers substrate

308 Ultra-Wideband Radio Technologies for Communications, Localization and Sensor Applications

however, it has to be taken into consideration.

proposed alternatives are not applicable.

*5.4.2. Measurement setup based on small antennas* 

following generalized formula:

algorithm [101]. *h*

weight coefficient [102].

preprocessing before beam-forming can be carried out. Most clutter removal approaches assume that the clutter appears very similar in each channel and, thus, its estimation improves with increasing channel number. It must be noted that this holds only for channels with comparable clutter parameters. That means clutter estimation and removal has to be done separately for groups consisting of only associated signals (channels with identical antenna distances and boresight angles Tx-Rx), which accomplishes this task. In scientific work on simulation, this circumstance is commonly ignored. For practical applications,

The simplest approach is to estimate the clutter by means of the average value. Tumor reflections are assumed to appear uncorrelated in the channels and to be negligible in the averaged signal. Even though publications about advanced clutter removal algorithms emphasize the weak points of this self-evident approach, it must be noted that it works relatively robustly in the case of covering tumor response by clutter when some of the

Image formation algorithms using time domain beam-forming can be included in the

where N is the number of channels, *<sup>n</sup> S t* is the clutter subtracted signal of channel *n* , <sup>0</sup> **r**

of channel *n* related to the focal point at 0**r** and *I***r**<sup>0</sup> is the back scattered energy which has to be mapped over the region of interest inside the breast. Based on two FIR filters, the different extensions of the common delay-and-sum beam former can be expressed. Pathdependent dispersion and attenuation [99], [100] can be equalized by means of <sup>0</sup> , *wn w*

which – in the simplest case - can be only a weight coefficient. Other improvements can also be included by convolution in the time domain, e.g. the cross-correlated back projection

The efficient penetration of the electromagnetic waves into the tissue and the spatial highresolution registration of the reflected signals are crucial tasks of the antenna array design. In this regard, efficiency is not only a matter of radiation efficiency or antenna return loss, respectively. An efficient antenna array design concerning biomedical UWB imaging purposes comprises also the shape and duration of signal impulses, angle dependence of the impulse characteristics (fidelity), and the physical dimensions of the antenna. These interacting parameters are hardly to accommodate to each other within one antenna design. Generally, compromise solutions have to be found considering basic conditions of scanning,

, ,( ) *h w*

0 0 0 0

 

symbolizes the coordinates of the focal point (image position vector), <sup>0</sup> ( ) *<sup>n</sup>*

2 12

*h h W w*

*T n T I h w St*

*T T N*

<sup>2</sup> 2 2

*h nw n n w h*

 **r r rr** (8)

*<sup>h</sup>* ,**r**<sup>0</sup> represents a smoothing window at the energy level or a scalar

 

**r** is the time delay

**r** Initially, we used short bow-ties (Fig. 48) with the dimensions of 8 mm x 3 mm implemented on Rogers® 4003 substrate (0.5 mm) using PCB technology. Dipoles have to be fed differentially. The balanced feeding is realized by differential amplifier circuits [103].

These antennas cannot be matched over a large bandwidth, which leads to unwanted reflections between antenna and amplifier. There are two options concerning the handling of this problem: realization of a sufficient line length between antenna and amplifier (in order to gate out the reflections) or implementation of the amplifier circuits directly at the antenna feed point. On an interim basis, we pursued the first strategy using long cables between antenna and amplifier. Assuming a maximum mean tissue permittivity ' 50 , a 70 cm cable will ensure that any reflections from inside of the breast (diameter ~ 10 cm) and unwanted reflections at the amplifier do not overlap.

As mentioned above, the contact between antennas and breast skin represents a crucial aspect for sufficient signal quality. Regarding clinical requirements (e.g. disinfection) we plan to place the antennas behind a thin examination mold. But this additional interface reduces the signal quality significantly. Therefore, a thin (~2 mm) matching layer consisting of tissue mimicking phantom material was inserted between the examination mold and the antennas in order to increase the signal energy penetrating the tissue and reduce the backward radiation (Fig. 49). The benefit achieved when using a thin contact layer was also investigated and verified by simulations (Fig. 5 in section 2.3.2).

We built up two preliminary array set-ups for phantom measurements, both including eight antennas and distributing them around a circular segment (diameter 9.5 cm) in steps of 22.5°. An array with a horizontal antenna arrangement is shown in Fig. 50. Exemplary phantom measurement results achieved with these prototypes are published in [104] and [105] and will be summarized in section 5.4.3.

310 Ultra-Wideband Radio Technologies for Communications, Localization and Sensor Applications

**Figure 49.** Schematic illustration of the contact layer filled with phantom material and mounted antennas inside

ultraMEDIS – Ultra-Wideband Sensing in Medicine 311

dielectric powders (e.g. carbon meal or barium titanate powder) will be admixed to silicone

**Figure 51.** Photographs of an active antenna element (Rx) with 8 mm dipole with amplifier circuit board (left panel) and the slide-on mounting system for phantom measurements as well as *in vivo*

The breast phantoms are tissue mimicking oil-gelatin phantoms according to [24] and described in section 5.2, where the dielectric properties can be adjusted by means of the oil content. For our measurements we used two types of material: 40% oil (57% water) content material mimics healthy tissue which approximately corresponds to group II of adiposedefined tissue (31%-84% adipose tissue) [106]. The 10% oil (85.5% water) content material simulates tumor tissue. Fig. 52 illustrates permittivity, attenuation losses and reflection coefficient between both tissues. In order to realize an optimal contact to the antenna array, the phantom material is filled in identical plastic containers (diameter 9.5cm) as used for the examination mold. The containers are hermetically sealed and stored in the fridge to avoid chemical instability of the phantom material. The phantoms have to be acclimatized at least

**Figure 52.** Dielectric values of the tissue mimicking phantom material: Permittivity (above),

transmission losses per cm and reflection coefficient between them (below)

rubber. This special challenging topic is currently under investigation.

measurements

*5.4.3. Imaging results of phantom trials* 

3 hours before starting the measurements.

**Figure 50.** Antenna array: Assembly stage before casting the contact layer. The connected and affixed differential fed antennas and the container for the outer boundary of the contact layer are still visible (left panel). Finished antenna array with inserted rotatable breast phantom (right panel)

After this preliminary development stage, the differential feeding amplifier was relocated into the antenna feed point. By this step, reflections due to antenna mismatch will be avoided, and the quantity of feeding cables will be bisected, because each active antenna element can be fed single-ended (Fig. 51).

In conjunction with this enhancement, the mechanical part of the antenna array was improved. A developed slide-in mounting system (Fig. 51) allows flexible antenna application and replacement and, therefore, facilitates investigations of various Rx-Txarrangements without destruction and rebuild of the whole array as it is the case with the preliminary set-up shown in Fig. 50 [106].

Because the contact layer will not be hermetically sealed in this case, the chemical instable oil-gelatin phantom material cannot be used anymore for this task. Thus, investigations of alternative materials have to be considered. We propose polymer-powder composites where

dielectric powders (e.g. carbon meal or barium titanate powder) will be admixed to silicone rubber. This special challenging topic is currently under investigation.

**Figure 51.** Photographs of an active antenna element (Rx) with 8 mm dipole with amplifier circuit board (left panel) and the slide-on mounting system for phantom measurements as well as *in vivo* measurements

### *5.4.3. Imaging results of phantom trials*

310 Ultra-Wideband Radio Technologies for Communications, Localization and Sensor Applications

**Figure 49.** Schematic illustration of the contact layer filled with phantom material and mounted

**Figure 50.** Antenna array: Assembly stage before casting the contact layer. The connected and affixed differential fed antennas and the container for the outer boundary of the contact layer are still visible

After this preliminary development stage, the differential feeding amplifier was relocated into the antenna feed point. By this step, reflections due to antenna mismatch will be avoided, and the quantity of feeding cables will be bisected, because each active antenna

In conjunction with this enhancement, the mechanical part of the antenna array was improved. A developed slide-in mounting system (Fig. 51) allows flexible antenna application and replacement and, therefore, facilitates investigations of various Rx-Txarrangements without destruction and rebuild of the whole array as it is the case with the

Because the contact layer will not be hermetically sealed in this case, the chemical instable oil-gelatin phantom material cannot be used anymore for this task. Thus, investigations of alternative materials have to be considered. We propose polymer-powder composites where

(left panel). Finished antenna array with inserted rotatable breast phantom (right panel)

element can be fed single-ended (Fig. 51).

preliminary set-up shown in Fig. 50 [106].

antennas inside

The breast phantoms are tissue mimicking oil-gelatin phantoms according to [24] and described in section 5.2, where the dielectric properties can be adjusted by means of the oil content. For our measurements we used two types of material: 40% oil (57% water) content material mimics healthy tissue which approximately corresponds to group II of adiposedefined tissue (31%-84% adipose tissue) [106]. The 10% oil (85.5% water) content material simulates tumor tissue. Fig. 52 illustrates permittivity, attenuation losses and reflection coefficient between both tissues. In order to realize an optimal contact to the antenna array, the phantom material is filled in identical plastic containers (diameter 9.5cm) as used for the examination mold. The containers are hermetically sealed and stored in the fridge to avoid chemical instability of the phantom material. The phantoms have to be acclimatized at least 3 hours before starting the measurements.

**Figure 52.** Dielectric values of the tissue mimicking phantom material: Permittivity (above), transmission losses per cm and reflection coefficient between them (below)

Figure 53 shows two measured signals of the proposed antennas which illustrate the appropriate time domain characteristics. The measurement through 6 cm tissue (mimicked by means of phantom material) as well as the cross-talk signal between two antennas show relatively short impulse shapes with low ringing, which is essential for UWB imaging. Including the dispersive tissue impact the spectral bulk ranges between 1 GHz and 3 GHz with a bandwidth greater than 2 GHz for both received impulses. Obviously, because of the dielectric scaling due to the direct contact between tissue and antenna, such small antennas are capable of radiating waves in a frequency range with acceptable attenuation and penetration depth.

ultraMEDIS – Ultra-Wideband Sensing in Medicine 313

relatively low dielectric contrast between healthy and cancerous tissue mimicking phantom

The results underline that small dipoles can be profitably applied for UWB breast imaging. The impressive identification of the tumor surrogates promises also the detection of weaker dielectric contrasts. On the other hand, it must be noted that the tumor surrounding tissue imitation is completely homogeneous which does not correspond to reality. Therefore, our breast phantoms must be enhanced in the future toward a better approximation of the breast

**Figure 54.** UWB images of phantom trials including a 15mm (top) and a 10 mm (middle) tumor surrogate as well as two 15 mm tumor surrogates, separated by 30 mm (below). Left: linear energy

material, the tumors can be detected and separated.

tissue heterogeneity.

scale; Right: logarithmic scale in dB.

**Figure 53.** Measurement signals based on the described bow-ties: measurement through 6 cm tissue mimicking phantom material with 40% oil content (left panel) and cross-talk signal between adjacent antennas, separated by 2.5 cm (right panel)

During the phantom measurements, four antennas acted as receivers and are permanently connected with Rx1…Rx4 of the radar device. The transmitter signal was connected to one of 4 transmitter antennas by a coaxial switch matrix. Thus, 16 signal channels could be achieved without rearrangement. Their angles between the boresight directions of Tx and Rx differed in the range 22.5 - 157.5°. Because this amount of signal channels is insufficient for high-resolution imaging, we had to consider robust and reproducible mechanical scanning to achieve a sufficient number of channels. In order to simulate antenna rotation, the phantoms were rotated in steps of 11.25°. This resulted in 512 signals (16 channels x 32 rotations) which could be included into the imaging process of one phantom.

Figure 54 shows exemplary imaging results of the described breast phantoms applying the presented measuring set-up and time domain beam-forming. Despite the relatively low dielectric contrast between both tissue simulations, the tumor inclusions can clearly be identified. The highest interferences (side lobes) are about 11dB (15mm tumor) and around 7dB (10mm tumor) lower than the tumor representation. Additionally, the lower panels of Fig. 54 illustrate the capability of localization and differentiation between multiple tumors, for example two 15 mm tumors with a distance of 30 mm between them. Despite of the relatively low dielectric contrast between healthy and cancerous tissue mimicking phantom material, the tumors can be detected and separated.

312 Ultra-Wideband Radio Technologies for Communications, Localization and Sensor Applications

penetration depth.

antennas, separated by 2.5 cm (right panel)

Figure 53 shows two measured signals of the proposed antennas which illustrate the appropriate time domain characteristics. The measurement through 6 cm tissue (mimicked by means of phantom material) as well as the cross-talk signal between two antennas show relatively short impulse shapes with low ringing, which is essential for UWB imaging. Including the dispersive tissue impact the spectral bulk ranges between 1 GHz and 3 GHz with a bandwidth greater than 2 GHz for both received impulses. Obviously, because of the dielectric scaling due to the direct contact between tissue and antenna, such small antennas are capable of radiating waves in a frequency range with acceptable attenuation and

**Figure 53.** Measurement signals based on the described bow-ties: measurement through 6 cm tissue mimicking phantom material with 40% oil content (left panel) and cross-talk signal between adjacent

During the phantom measurements, four antennas acted as receivers and are permanently connected with Rx1…Rx4 of the radar device. The transmitter signal was connected to one of 4 transmitter antennas by a coaxial switch matrix. Thus, 16 signal channels could be achieved without rearrangement. Their angles between the boresight directions of Tx and Rx differed in the range 22.5 - 157.5°. Because this amount of signal channels is insufficient for high-resolution imaging, we had to consider robust and reproducible mechanical scanning to achieve a sufficient number of channels. In order to simulate antenna rotation, the phantoms were rotated in steps of 11.25°. This resulted in 512 signals (16 channels x 32

Figure 54 shows exemplary imaging results of the described breast phantoms applying the presented measuring set-up and time domain beam-forming. Despite the relatively low dielectric contrast between both tissue simulations, the tumor inclusions can clearly be identified. The highest interferences (side lobes) are about 11dB (15mm tumor) and around 7dB (10mm tumor) lower than the tumor representation. Additionally, the lower panels of Fig. 54 illustrate the capability of localization and differentiation between multiple tumors, for example two 15 mm tumors with a distance of 30 mm between them. Despite of the

rotations) which could be included into the imaging process of one phantom.

The results underline that small dipoles can be profitably applied for UWB breast imaging. The impressive identification of the tumor surrogates promises also the detection of weaker dielectric contrasts. On the other hand, it must be noted that the tumor surrounding tissue imitation is completely homogeneous which does not correspond to reality. Therefore, our breast phantoms must be enhanced in the future toward a better approximation of the breast tissue heterogeneity.

**Figure 54.** UWB images of phantom trials including a 15mm (top) and a 10 mm (middle) tumor surrogate as well as two 15 mm tumor surrogates, separated by 30 mm (below). Left: linear energy scale; Right: logarithmic scale in dB.
