Safety assessment of ultra-wideband antennas for microwave breast imaging

Safety Assessment of UWB Exposures

The sensors used for microwave breast imaging are typically placed close to the breast, necessitating calculation of energy absorptions inside the breast. However, due to the pulsed nature of UWB technology, this calculation and comparison with safety limits is complicated.

Firstly, there are only a few reports (to the knowledge of the authors) pertaining to UWB exposures. The safety assessment of UWB radio systems for body area network (BAN) applications was investigated by Wang and Wang [2008] and De Santis et al. [2010]. Also, two reports dealt specifically with UWB exposures expected during microwave breast imaging [Zastrow et al., 2006, 2007]. Both articles numerically analyzed realistic models of the breast when illuminated with a ridged pyramidal horn antenna purposely designed for breast imaging. The first article describes the safety assessment at selected frequencies [Zastrow et al., 2006], while a time-domain approach is discussed by Zastrow et al. [2007]. As these reports are specific to a particular antenna, there is a need for safety evaluation of other designs proposed for microwave breast imaging application.

Secondly, the ICNIRP and IEEE C95.1 safety standards [ICNIRP, 1998; IEEE Std C95.1, 2006] are not clear on exposure to UWB fields. Indeed, at microwave frequencies, the most effective dosimetric quantity adopted by such standards is the specific absorption rate (SAR). The limits on SAR are established to prevent heating effects. However, this basic restriction is meaningful only for time-harmonic fields. For pulsed fields, exposure limits are instead given in terms of specific absorption (SA) to prevent the microwave hearing effect, and temporal peak of the in situ electric field to prevent electrostimulations. Nevertheless, in the opinion of the authors, both pulsed restrictions are not easily applied to microwave breast imaging exposures since the first effect concerns localized head exposures, while the second effect occurs at low frequencies. Therefore, for UWB microwave breast imaging, it is not clear which is the most meaningful dosimetric quantity. The SA is likely an appropriate and effective measure, however, the SA limits to be adopted by safety standards differ in terms of averaging time. Specifically, a peak value of 28.8 J/kg averaged over 10 g should never be exceeded over any one-tenth second period [IEEE Std C95.1, 2006]. It is worthy noting that such a value is derived from maximum permissible exposure referred to whole-body occupational exposures (i.e., 0.4 W/kg) and therefore a higher value of 144 J/kg should be applied when considering localized exposures for general public (i.e., 2 W/kg). A value of 2 mJ/kg averaged over 10 g is instead specified for the general public by the ICNIRP [1998] safety guideline. In this case, the averaging time is not clearly specified (see Table 4, note 7 in ICNIRP [1998]). However, it is explicitly claimed that microwave hearing occurs for pulses of duration less than 30 {micro}s and therefore it is reasonable to consider the same order of magnitude for the SA averaging time.

Finally, a further critical issue concerns the fact that microwave breast imaging systems are still in a prototypal phase. This means that, for practical reasons, a multi-frequency vector network analyzer (VNA) is typically used to collect measurements. The breast is illuminated by sweeping over a frequency range and collecting measurements at selected frequencies, rather than by direct time-domain measurements of a UWB pulse. Thus, the time-domain SA evaluation employed by Zastrow et al. [2007], Wang and Wang [2008], and De Santis et al. [2010] cannot be applied here. Therefore, a novel frequency-domain SA evaluation approach starting from the SAR values obtained at different frequencies is proposed in this article.

Antenna Models

We have developed three different antennas for use in prototype microwave breast imaging systems at the University of Calgary, Alberta, Canada (Fig. 1). The first two sensors are based on the well-known BAVA. Both Vivaldi antennas have been designed to operate over the frequency range from 2 to 12 GHz, although measurements may occur over a wider frequency range. The antennas are placed at a distance of 1 cm from the breast, and in an immersion liquid of canola oil. The canola oil permits reduction of both the sensor size and reflection from the skin when compared to free space. The second BAVA includes a dielectric director (BAVA-D), which is a novel feature introduced to additionally focus the near-field radiation in the endfire direction. Both versions of the antenna are thoroughly described in Bourqui et al. [2010b].

The third antenna is a tapered slot antenna fed by a transverse microstrip balun and it also includes a dielectric director. This antenna is termed the Cassiopeia antenna. The Cassiopeia antenna is designed to come in contact with the breast skin, and also to operate in a variety of immersion media (i.e., relatively independent of immersion medium). The antenna reported in this article is tested in an immersion liquid of pure glycerin. A detailed description of the Cassiopeia antenna is provided in Bourqui et al. [2010a].

Breast Model

EM Numerical Modeling

All the simulations are performed with the finite-difference time-domain (FDTD) technique using the commercial software tool, SEMCAD X (SPEAG). The several combinations of antennas and breast models are excited by a modulated Gaussian pulse with frequencies in the range of 1–15 GHz, which is the typical frequency band for radar-based microwave imaging techniques.

Following the experimental procedure, the BAVA and BAVA-D antennas are placed 1 cm from the breast (metallization end as reference), while the Cassiopeia is in contact with the breast model (Fig. 2). Moreover, the breast models illuminated by Vivaldi antennas are placed in canola oil, while those using the Cassiopeia antenna are placed in pure glycerine. The computational domain is truncated by perfectly matched layers (PMLs) [Bérenger, 2003].

Finally, the tissue dielectric properties are represented by Debye dispersion models in order to account for the frequency dependence inherent in UWB pulse propagation. Indeed, the simple Debye model represents a fair approximation of the more complex Cole–Cole dispersive model in the limited frequency range of 1–15 GHz [Lazebnik et al., 2007b]. Table 1 summarizes the Debye parameters for the cylindrical model. In particular, dielectric properties of the skin are obtained from Gabriel [1996], while those of the homogenous interior tissue are from Lazebnik et al. [2007a, b]. The dielectric properties of the 16-tissue realistic breast model are assigned electrical properties based on Lazebnik et al. [2007a, b] by mapping the pixel intensity to the electrical properties. In particular, a histogram of the breast model is created prior to segmentation. Then, it is determined that voxels with <30% of the maximum pixel intensity are considered fatty tissue, while those greater than 85% of the maximum pixel intensity are considered glandular tissue. The voxels that are related to the remaining pixels are termed transition regions. These tissue types correspond to the general groups defined in Lazebnik et al. [2007a]. For each tissue type, the segmented regions are mapped to the properties in the 25th and 75th percentiles for each tissue that is defined in Lazebnik et al. [2007a]. It should be noted that the average dielectric properties of the interior tissues of the real breast model are close to those of the cylindrical model with Group 2 material assigned to the interior.

Safety Assessment

Antenna Validation Method

As a first step, we measure and simulate SAR for each of the antennas of interest in a simple scenario and at a single frequency. This configuration is aimed at ensuring that measurements and simulations for each antenna are in agreement in order to validate the adopted radiating system.

To assess the antennas, a custom phantom setup has been implemented, as shown in Figure 3. The setup includes a 45.7 cm diameter cylindrical tank filled with canola oil. The antenna of interest is placed at the center of the bottom of the tank, pointing up. At the tip of the antenna is a square box filled with a head-simulating liquid (HSL) used to mimic the electrical properties of the head at 2.45 GHz (i.e., ε_r = 36.7 and σ = 1.89 S/m).

Measurements are performed with a DASY4 professional system (SPEAG), which is normally used in SAR measurements for compliance testing. The DASY system has an E-field probe that is used to measure the electric field inside the box of HSL by scanning the probe through a region of interest. The measurement protocol used with the DASY system follows the recommendations of the IEEE Std C95.3 [2002].

The simulations of the measurement system are aimed at reproducing the results from the measurements. Therefore, we created a replica of the measurement system by including all parts of the measurement setup except for the E-field probe. In both cases the SAR is evaluated by:

(1)

where ρ is the tissue mass density at spatial location r, and f is the considered frequency.

SA evaluation method

The SA produced inside the breast by any pulsed microwave imaging technique can be numerically evaluated as:

(2)

where T_max is the pulse duration, and J(r,t) is the current density vector inside the breast. An efficient way to numerically evaluate Equation (2) in dispersive media is described in detail in De Santis et al. [2010]. However, for actual prototype systems based on frequency-swept VNA techniques, the value of T_max even for a single sweep is significant (e.g., 1.8 s for TSAR system), making the safety assessment impractical from a numerical point of view. To overcome this problem, a novel procedure based on a frequency-domain approach has been proposed. Mimicking the measurement procedures, the SA produced by a single VNA sweep is numerically evaluated by:

(3)

where N is the number of frequency points (1601 for TSAR system), SAR(r, f_i) is the SAR produced at frequency f_i, and T_i is the time duration of the VNA sweep at frequency f_i (3.372 ms for a TSAR system).

In this case, a broadband signal excitation with limited time duration (i.e., on the order of nanoseconds) can be used to extract the SAR values at selected frequency points. For practical reasons, the SAR produced by the several antennas inside a selected breast model is numerically evaluated at only a few frequency points (e.g., every 1 GHz). After that, the data are interpolated to account for the missing SAR values (the frequencies at which measurements are collected) and normalized to a power of −5 dBm. The SAR values are then multiplied by the time that each frequency is radiated during a sweep (i.e., T_i = 3.372 ms). By summing the contribution at each frequency point, one obtains the SA per sweep. Finally, by knowing the number of sweeps that are performed for a given time, one can extrapolate the total SA resulting from a 6 min exposure or a complete TSAR scan (i.e., 30 min). Specifically, 120 sweeps are performed in 6 min while a total TSAR scan takes 600 sweeps. A validation of the proposed technique, together with a sensitivity analysis on the interpolation effect, is reported in the following section.

Antenna Validation

To examine the results of measurements and simulations at a single frequency we present the SAR values, which include the 10 g and unaveraged peak SAR values. The measurement and simulation results in Table 2 show that the SAR values for measurements and simulations are in good agreement. We note that the BAVA-D values are smaller than the values for the BAVA due to differences in antenna positioning. For the BAVA, the antenna is positioned so the end of the metallization is 1 cm from the box of HSL. With the BAVA-D, the antenna is positioned so the end of the director is 1 cm from the box. Therefore, the aperture (end of metallization) of the BAVA-D is located further from the HSL than the BAVA, resulting in smaller values.

These results demonstrate that our measurements and simulations are in agreement at a single frequency of 2.45 GHz. Figure 4 compares simulated and measured reflection coefficients for the three antennas, suggesting that we can extend the validity of the radiating system over the frequency range of interest.

SA Validation

To validate the proposed SA frequency-domain approach, a simple test case is considered, as shown in Figure 5a. It consists of a homogenous cubic tissue with Debye parameters (ε_s = 56, ε_∞ = 26, σ = 0.64 S/m, and τ = 0.17 ns) excited by a plane wave with two different chirped signals (Fig. 5b). The first signal (called s₁) is a three-tone harmonic signal with frequencies f₁ = 0.6 GHz, f₂ = 0.7 GHz, and f₃ = 0.8 GHz, while the second chirped signal (called s₂) has frequencies at f₁ = 3 GHz, f₂ = 4 GHz, and f₃ = 5 GHz. Both signals have the same time duration T_max = 3 ns with a constant single-tone time window of T_i = 1 ns. The SA produced inside the cubic tissue along the x-axis is shown in Figure 6 for both time- and frequency-domain approaches. As can be observed, the correlation is quite poor for the first chirped signal (average error of 20%), while good agreement is obtained for the second signal (average error of 5%). This means that Equation (3) is valid only when the considered harmonic tone f_i has an even number of cycles (or reaches steady-state) within the considered time window T_i. It can be concluded that the adopted approach is consistent for TSAR safety assessments due to the long sweep time compared to the high involved frequencies. Indeed, even for the minimum illuminating frequency of 1 GHz, T_i is on the order of milliseconds and therefore the steady-state is always satisfied.

TSAR Exposure

As previously described, the SA is obtained by means of a frequency-domain procedure starting from the SAR values evaluated at different frequencies. For the sake of completeness, first the SAR distribution at selected frequencies is illustrated. Then, the effect of the SAR interpolation on the SA evaluation is investigated. Figure 7 shows the SAR distribution (unaveraged) produced by the BAVA antenna inside the cylinder breast model (filled with material representing Group 1) at four selected frequencies when adopting a power output of −5 dBm at each frequency. In order to highlight this absorption effect and improve the picture readability, the SAR distributions are represented in a logarithmic scale by normalizing the maximum value (0.0625 W/kg) to 0 dB. From this figure it is evident that the skin depth phenomena lead to a stronger but superficial absorption as the frequency increases. The analysis of the interpolation effect is reported in Figure 8, where the SA maps (unaveraged and logarithmic scale) obtained inside the realistic breast model when illuminated by the BAVA antenna are shown for different numbers of extracted frequency points N. Specifically, the SAR values were extracted at frequencies f_i = 1, 5, 10, 15 GHz for N = 4 (Fig. 8a); f_i = 1, 2, … , 14, 15 GHz for N = 15 (Fig. 8b); and f_i = 1, 1.5, 2, … ,14, 14.5, 15 GHz for N = 29 (Fig. 8c). Directly following the TSAR measurement procedure and extracting the SAR values at 1601 frequency points is computationally intensive and not practical. Although N = 4 appears to be an insufficient number of frequency points, the distributions at N = 15 and N = 29 are similar and the difference in the SA values is <2%. This suggests extracting the SAR values at only 15 frequency points and interpolating the remaining 1586 points, as shown in Figure 8d. In the following, all the SA results have therefore been evaluated by means of this interpolation procedure, representing a good trade-off between accuracy and computational cost.

The SA distributions (unaveraged and logarithmic scale) obtained inside the realistic breast model for the three antennas resulting from exposure to a single VNA sweep are reported in Figure 9. As expected, the peak SA is concentrated close to the skin due to the high conductivity of this layer and the relatively high frequency involved. When comparing the BAVA and BAVA-D results, increased concentration of the distribution is noted for the BAVA-D, consistent with the increased focus of the fields due to the director. With the Cassiopeia, greater penetration of the fields is evident due to the contact between this antenna and the breast.

In addition to the SA/SAR distributions, it is relevant to quantify the absorptions and compare them with safety limits. The results of our investigation are therefore summarized in Tables 3 and 4. In particular, the specific absorptions obtained inside the breast models after exposure to a single sweep by the several UWB antennas are listed in Table 3, together with the SAR values at selected frequencies. The SA extrapolation after a 6 min exposure and a total TSAR scan are reported in Table 4 for the realistic MR breast model. It should be noted that these last values are overestimated since different regions of the breast are illuminated as the antenna location varies during the image scan; in turn, the SA peak location also changes position. Thus, multiplying the single SA per sweep by the number of sweeps performed within a specified time window represents the worst-case scenario of a fixed antenna illuminating the breast.

The analysis of the obtained results indicates that the highest absorptions are obtained for the Cassiopeia antenna (since it is in contact with the breast). The BAVA-D results are very similar to the Cassiopeia due to the focusing effect of the director in the BAVA-D. For the realistic breast models, the 10 g SA values for the BAVA and BAVA-D are between the cylinder Group 1 and Group 2 results, which is reasonable given the dielectric properties of the models. For the Cassiopeia antenna, the contact with the cylindrical model is more consistent than the contact with the irregular shape of the realistic breast model. This leads to higher absorptions in the cylinder. On the other hand, due to the tissue composition and shape variation from patient to patient, the results in Table 3 suggest that the cylindrical breast models represent meaningful testbeds from a safety point of view. In other words, the cylindrical Group 1 model is useful for evaluating the worst-case scenario with the highest energy absorptions.

Finally, comparing the results with the safety limits suggests that no particular concerns are anticipated. The results in terms of SAR are well below the safety limit of 2 W/kg averaged over 10 g as specified by the ICNIRP and IEEE safety standards. As already mentioned, the choice of the most suitable dosimetric quantity for UWB microwave imaging is not clear. By assuming that SA is the most meaningful quantity, then the ICNIRP standard states a SA limit of 2 mJ/kg over 10 g. This value is usually exceeded in our calculations for a single pulse (Table 3); however, the averaging time in the standard likely refers to the microwave hearing pulse duration (i.e., microseconds), while the considered VNA sweep is on the order of seconds (i.e., 1.8 s). An unclear scenario is found once again when considering the SA limits provided by the IEEE safety standards. For instance, Zastrow et al. [2007] adopted the SA limit of 576 J/kg over 1 g for a 6 min exposure, but this limit referred to a previous version of the standard [IEEE Std C95.1, 1999]. However, even considering the new value of 144× J/kg over 10 g and any 0.1 s period [IEEE Std C95.1, 2006], the calculated energy absorptions are well below the SA limit suggested by the IEEE safety standard.