3. Digital breast tomosynthesis (DBT)

the age range of 50–51 in their first mammography session. In the following sessions, the probability was approximated to be 3.8%; but the cumulative risk of getting an FP outcome turned out to be 32.4%, implying that one out of three women could receive an FP alarm during a 10-year biannual screening program [21]. The large study performed in 2011 reported that there is a prevalence of 16.3 and 9.6% in obtaining an FP result in the first and following sessions, severally. The cumulative risk was found to be 61.3% for women starting their regular session at the age of 40 or 50, in an annual screening program [22]. Before the possibility of breast cancer is ruled out, a patient who receives a positive result is recalled for further examination that usually includes ultrasound imaging, needle biopsy, and sometimes surgical biopsies. These biopsies, which some authors call "unnecessary" [23, 24], as well as the prolonged uncertainty about ones' health status, can be accompanied by a psychological trauma similar to those who were diagnosed with cancer in their first 6 months with symptoms such as short-term anxiety, more frequent self-examinations, and a change in the patients' tendency to attend later mammography sessions [25–28]. It has also been reported that 27 and 33% of women gone under breast biopsy had reduced sexual sensitivity and pain in the breast. In addition, an increment in mood disturbances and reduction in natural killer cell activity and INF γ production was observed before and after biopsy [26, 27]. Therefore, FP results can lead

As the name implies, when a malignant tissue is not detected in a mammogram, the result is a false-negative (FN). The tumor might be detected in the next screening sessions, when perhaps it has grown visible due to invasion into the encompassing tissues. The data on the prevalence of FN results has been inconsistent. In a 1996 study, it was found that, depending on family history, follow-up duration, and age; between 12.5 and 31.5% of mammograms were interpreted as cancer-free, while there was actually a tumor in the breast [29]. A more recent study, in 2014, showed that FN probability in digital and screen-film mammography for women above (under) 50 was 24% (27%) and 27% (24%), respectively [30]. It was reported that obtaining an FN result was directly associated with breast density. Other factors that play a role in FN evaluation of breast are the type of cancer (particularly BRCA 1 and BRCA

Overdiagnosis may be defined as "the detection of cases that would never have come to clinical attention without screening [in a patient's lifetime] [32]." Though some physicians disagree with the numbers reported for such cases [33]. Some studies from 2004 reported overdiagnosis rates of 5% in ductile carcinoma in situ (DCIS) in Italy [33] and 33% in general mammography of women of ages 50–59 in Norway and Sweden [34]. The results from the later study were comparable with the 30% rate of overdiagnosis reported in [35], back in 2001. A larger study in 2005, which was carried out over eleven counties in Sweden, reported the risk of overdiagnosis as 21 and 54% for women in age groups of 50–59 and 60–69, in order [37]. The prevalence of subclinical tumors among women after 50 was explained as the reason for rather high risks of overdiagnosis. On the other hand, a recent report in 2015 from South

2 mutations) and the location of the tumor (being close to the chest wall) [31, 32].

to serious consequences that need to be addressed.

2.2. False-negative results

18 New Perspectives in Breast Imaging

2.3. Overdiagnosis

About 40% of eligible women who are recommended to have regular screening have dense breasts, which are composed of more than 50% glandular tissue and are among the major contributors to uncertain mammogram readings [38, 39]. Thereby, one of the focuses of new screening devices is to enable better imaging visibility through different layers of breast. This is not properly addressed in conventional 2D mammograms of the breast; since an existing tumor can be masked by underlying and overlying tissues and, thus, make accurate diagnosis difficult for radiologists. In order to enhance the image resolution, full-field digital mammography (FFDM) detectors can be utilized. The DBT employs these detectors as well as a rotating X-ray tube that moves in a circular arc to illuminate the breast from different angles [40]. Then, images of several 2D projections at dissimilar depths are reconstructed, making it feasible for radiologists to look through different layers. Aside from hardware enhancements, novel imaging algorithms –including, but not limited to, filtered back projection and Gaussian frequency blending; and iterative techniques such as Maximum Likelihood and Simultaneous Algebraic Reconstruction Technique—have been applied in the DBT on the software level [40].

Conventional mammography makes use of film-screening, which provides high resolution to delineate micro-calcifications and other fine features in the breast; yet, it has a narrow dynamic range that limits tumor visualization in dense segments of the breast. FFDM, however, improves dynamic range and provides the radiologists with the option to manipulate the reconstructed images. Moreover, higher contrast is observed between dense glandular and fatty tissue when FFDM detectors are used [40]. The efficacy of the DBT has been put to the test to evaluate how effective its new features are in practice. In a study in 2005, it was reported that for the age group under 50, premenopausal women, and patients with consistently or extremely dense breasts; the accuracy of images acquired by digital mammography was higher than that of conventional mammography [41]. Additionally, it was shown that the margin, calcification, and lesion visibility of the DBT was superior to conventional imaging in the images of an FDA-certified breast phantom. The radiation exposure levels were similar or less in the DBT; and the tumor boundaries and vessels around calcifications were clearer, which contribute to better identification of malignant and benign masses [42]. Based on the better lesion visibility of the DBT, Teersa et al. recommended tomosynthesis as a complementary modality to 2D mammography [42]. The combination of 3D and 2D mammography is more commonly adopted than 3D imaging alone in clinical settings [40].

The improved accuracy using 2D-3D mammography was also observed by Bernardi et al. [43, 44]. Better visualization of tumors, as provided by a 2D-3D setup could reduce the number of FP results and consequently result in fewer follow-up biopsies. Consistent with this prediction, in 2009, Gur and others reported 10% decrease in false-positives when DBT was used in addition to FFDM [45]. In a more recent study in 2014, the ascendency of 2D-3D mammography over 2D mammography was substantiated in terms of true-positive evaluation and sensitivity [46]. In the same year, Bernardi's subsequent study showed a considerable elevation, from 60% in 2D to 87% in true-positive readings of radiologists when using a 2D-3D configuration [47]. One year later, Svahns et al. considered the FP to TP ratios in three populationbased studies and reported improvements in radiologists' interpretive efficacy of X-ray images when a combined configuration was implemented. The results indicated 55, 48 and 30% improvement in true positive detection when a 2D-3D setup was used in lieu of traditional setup in Houston, STORM, and Oslo studies, severally [48]. In TOMMY trial in UK, 2015, increased specificity (by 9%) for all studied groups and elevated sensitivity (by 7%) for dense breasts was observed as a result of using digital and film screening [49]. A 2016 review also reported fewer recall rates using the DBT [50].

### 4. Breast near-field radar imaging (NRI)

#### 4.1. Background

The contrast between cancerous and fibroglandular tissues is higher when they are exposed to microwave radiation than when they are illuminated by X-ray [53]. This contrast is increased about 5–10 fold when moving form X-ray to microwave frequencies, contingent upon the operating frequency span [51, 52], which renders microwave imaging a distinct candidate to be used as a complementary modality to 2D-3D mammography. A large study by Lazebnik et al. showed that the difference between dielectric properties of malignant and normal tissue is 10% when the fibroglandular tissue is in the background and 10:1 when the adipose-dominant tissue is considered as reference [53]. Though the former is significantly lower than the adipose/tumor contrast, it is still remarkably higher than the fibroglandular/tumor contrast in Xray frequencies, which is about 1% [54]. Figure 2 displays the ex vivo tissue properties as reported in [53] and as can be observed, the dissimilarity between the fibroglandular and malignant tissue is appreciable. Halter et al. showed that the difference between the dielectric properties of healthy and malignant tissues measured in vivo, using microwave and electrical impedance spectroscopy during reduction surgeries, is decreased [54]. However, the contrast is still adequately large for NRI to be a technique of interest as an improvement to conventional or digital mammography. Meaney et al. also pointed out that even though the in vivo measurements exhibit lower microwave contrast of about 2:1 (tumor permittivity: average permittivity of breast) at 0.9 GHz, the difference is still significant compared to that in most other traditional imaging systems [52]. Besides high complex permittivity values, there are other features that Near-Field Radar Microwave Imaging as an Add-on Modality to Mammography http://dx.doi.org/10.5772/intechopen.69726 21

better lesion visibility of the DBT, Teersa et al. recommended tomosynthesis as a complementary modality to 2D mammography [42]. The combination of 3D and 2D mammography is

The improved accuracy using 2D-3D mammography was also observed by Bernardi et al. [43, 44]. Better visualization of tumors, as provided by a 2D-3D setup could reduce the number of FP results and consequently result in fewer follow-up biopsies. Consistent with this prediction, in 2009, Gur and others reported 10% decrease in false-positives when DBT was used in addition to FFDM [45]. In a more recent study in 2014, the ascendency of 2D-3D mammography over 2D mammography was substantiated in terms of true-positive evaluation and sensitivity [46]. In the same year, Bernardi's subsequent study showed a considerable elevation, from 60% in 2D to 87% in true-positive readings of radiologists when using a 2D-3D configuration [47]. One year later, Svahns et al. considered the FP to TP ratios in three populationbased studies and reported improvements in radiologists' interpretive efficacy of X-ray images when a combined configuration was implemented. The results indicated 55, 48 and 30% improvement in true positive detection when a 2D-3D setup was used in lieu of traditional setup in Houston, STORM, and Oslo studies, severally [48]. In TOMMY trial in UK, 2015, increased specificity (by 9%) for all studied groups and elevated sensitivity (by 7%) for dense breasts was observed as a result of using digital and film screening [49]. A 2016 review also

The contrast between cancerous and fibroglandular tissues is higher when they are exposed to microwave radiation than when they are illuminated by X-ray [53]. This contrast is increased about 5–10 fold when moving form X-ray to microwave frequencies, contingent upon the operating frequency span [51, 52], which renders microwave imaging a distinct candidate to be used as a complementary modality to 2D-3D mammography. A large study by Lazebnik et al. showed that the difference between dielectric properties of malignant and normal tissue is 10% when the fibroglandular tissue is in the background and 10:1 when the adipose-dominant tissue is considered as reference [53]. Though the former is significantly lower than the adipose/tumor contrast, it is still remarkably higher than the fibroglandular/tumor contrast in Xray frequencies, which is about 1% [54]. Figure 2 displays the ex vivo tissue properties as reported in [53] and as can be observed, the dissimilarity between the fibroglandular and malignant tissue is appreciable. Halter et al. showed that the difference between the dielectric properties of healthy and malignant tissues measured in vivo, using microwave and electrical impedance spectroscopy during reduction surgeries, is decreased [54]. However, the contrast is still adequately large for NRI to be a technique of interest as an improvement to conventional or digital mammography. Meaney et al. also pointed out that even though the in vivo measurements exhibit lower microwave contrast of about 2:1 (tumor permittivity: average permittivity of breast) at 0.9 GHz, the difference is still significant compared to that in most other traditional imaging systems [52]. Besides high complex permittivity values, there are other features that

more commonly adopted than 3D imaging alone in clinical settings [40].

reported fewer recall rates using the DBT [50].

4.1. Background

20 New Perspectives in Breast Imaging

4. Breast near-field radar imaging (NRI)

Figure 2. Comparison between the dielectric properties and conductivity of healthy tissues and malignant tissues. Plots are based on the data from [53].

make NRI a suitable candidate for breast screening. It is intrinsically non-ionizing and it does not require breast compression when used independently. Also, microwave imaging provides high sensitivity for revealing small-sized masses. These qualities come at a considerably lower cost than other methods such as nuclear medicine and magnetic resonance imaging (MRI) [55].

To be able to reconstruct images of breast in microwave frequencies, several studies have been dedicated to find practical imaging algorithms including [56–58] that introduce finite difference time-domain (FDTD) and finite element techniques. Among other introduced methods, iterative numerical techniques such as conjugate gradient least square, Newton-Kantorovish, and Levenberg-Marquardt [59–61]; as well as finite difference frequency-domain (FDFD) can be mentioned [62–67]. The moderate imaging resolution in current clinical layouts can be enhanced via novel algorithms such as solving the 3D inverse problem employing variablestrength spatial prior constraints [66] and applying compressive sensing confocal imaging algorithms [67]. Both of these methods present encouraging results for the future of NRI. Making use of prior geometry or information, acquired by other imaging modalities, for microwave imaging is an ongoing research subject that can facilitate breast tomography and image reconstruction. Two recent works in this area used microwave radar imaging and MRI to feed the dielectric parameter reconstruction codes with spatial information. Baran et al. restored the average dielectric properties of different tissue types by microwave radar imaging and then used the data to perform microwave tomography [68]. They were able to obtain finer reconstruction qualities using this hybrid method than when tomography and radar-based imaging were conducted separately. Golnabi et al. employed an MRI image to reconstruct the complex permittivity of a real breast [69]. They reported improved accuracy and contrast between the target and background medium that were fibroglandular and adipose tissues, respectively. One year later, in 2016, they implemented this idea in a 3D reconstruction and again achieved promising outcomes with up to 2 and 9 times improvement in accuracy of conductivity and permittivity maps, respectively [70]. Due to the interference of metallic parts with the magnetic fields, combining microwave imaging and MRI in the same physical setup was reported to be unfeasible.

#### 4.2. Breast NRI at experimental level

One of the first NRI system implementable for a clinical trial was introduced in 2000 by Meaney and his colleagues [52]. The operational bandwidth of the system was 0.3–1 GHz and it was configured in a way that a patient could comfortably lie down on a bed with her breast pendent in a liquid inside a container with a circular antenna array encompassing the breast. The antennas were immersed in the same liquid (saline) as the matching medium, and they collected the scattered field from the submerged breast from multiple directions. The scan took 10–15 min for each breast of the human participants of the trial, using a tomographic method from the chest to the nipple. As predicted, the reconstructed maps of dielectric properties were of low resolution; however, they showed that the in vivo breast properties are larger than the ones obtained from ex vivo measurements. Later in 2007, the same group reported a good agreement between images retrieved by microwave imaging and MRI from subjects with normal mammograms and multiple phantoms [66]. Particularly, it was shown that fibroglandular tissue distribution and water content agreed well between the two sets of images. Furthermore, they found that similar to fat percentage and density, dielectric properties of the breast are heterogeneous.

Klemm et al. assembled a similar physical setup except, in place of a circular antenna, they used a set of ultra-wideband (UWB) antennas that were arranged in a hemispherical shape to conform to the shape of the pendent breast [71]. The system's performance was evaluated based on experiments carried out on a phantom encapsulated in a shell mimicking skin. They were able to detect small tumor simulants, 4-6 mm in diameter, embedded in the phantoms by implementing two beamforming algorithms, multi-static microwave imaging (MAMI) and delay-and-sum (DAS). Later, the system was tested on highly heterogeneous phantoms and it was shown to give promising results [72]. Also, in a clinical case study, the system was tested on a real breast, and the obtained image was compared to its associated mammogram. The malignant tissue was correctly located in the microwave image when compared to the X-ray image [73].

Lai and his group also employed UWB antennas to scan a number of heterogeneous and homogenous breast phantoms at Nanyan Technological University [74]. By rotating the phantoms 360 degrees, the antennas simulated a circular set of 360 elements. This configuration enabled them to detect a 4 mm tumor in all phantom types. But, detectability was not strong enough to delineate 2 mm inclusions. Lazero et al. published the results of what they named a "worst-case-study" utilizing a single UWB monopole antenna [75]. The phantom model was filled with water (presenting high losses), with a rod submerged inside it, and the antenna was located in air (presenting high reflection). Notwithstanding these challenging conditions, the embedded rod was successfully detected, implementing FDFD and Wiener-filter algorithm that took into account the effect of a simulated skin interface. Chun Yu and colleagues, in a different experimental layout, used an active microwave imaging method to retrieve images of a clay ball, a metallic ball, and a combined arrangement of both when they were submerged in water [76]. They utilized a dipole single transmitter and receiver to avert the mutual coupling that exists in an antenna array. The transmitting and receiving array were moved by an automatic positioner to enable a flexible, multi-view data acquisition in three dimensions over the surface of the object being measured. Applying an active, hybrid-inversion, reconstruction method, they were able to image one clay ball and the combined arrangement of two clay balls and one metallic ball.

Henriksson et al. made use of a microwave camera, at 2.45 GHz, to quantitatively restore the dielectric properties of materials [77]. In a quantitative approach, in place of approximations, the non-linear diffraction problem is solved at the cost of heavy computations. In 1990, the planar camera was built for non-invasive hyperthermia control; and later, in 1998, it was put to the test by Franchois on a rotating object inside a liquid container. Henriksson and his colleagues used a similar experimental setup that included two horn antennas. They made a breast-equivalent liquid using a Triton X-100 mixture and placed it inside a rotating tube—the object of interest—to have a more realistic model. Albeit some artifacts were present, the system performed well in reconstructing dielectric maps, using the Newton-Kantorovich algorithm. This camera enables NRI to be carried out in a mammographic configuration [77, 78], one of the key benefits of the approach presented in this chapter. More recently, in 2013, imaging a moving target was also tested in an integrated microwave imaging radar, as in an inverse synthetic array radar, at the University of Padova in Italy [79]. In the assembled setup, a planar antenna, consisting of two monopoles and two feed lines, was fixed on a support frame and it radiated down over the moving phantom. Two stepper motors shifted the target in the horizontal plane in two perpendicular directions. The results demonstrated that the system is capable of detecting two embedded inclusions inside the phantom.

4.2. Breast NRI at experimental level

22 New Perspectives in Breast Imaging

One of the first NRI system implementable for a clinical trial was introduced in 2000 by Meaney and his colleagues [52]. The operational bandwidth of the system was 0.3–1 GHz and it was configured in a way that a patient could comfortably lie down on a bed with her breast pendent in a liquid inside a container with a circular antenna array encompassing the breast. The antennas were immersed in the same liquid (saline) as the matching medium, and they collected the scattered field from the submerged breast from multiple directions. The scan took 10–15 min for each breast of the human participants of the trial, using a tomographic method from the chest to the nipple. As predicted, the reconstructed maps of dielectric properties were of low resolution; however, they showed that the in vivo breast properties are larger than the ones obtained from ex vivo measurements. Later in 2007, the same group reported a good agreement between images retrieved by microwave imaging and MRI from subjects with normal mammograms and multiple phantoms [66]. Particularly, it was shown that fibroglandular tissue distribution and water content agreed well between the two sets of images. Furthermore, they found that similar to fat percentage and density, dielectric properties of the breast are heterogeneous.

Klemm et al. assembled a similar physical setup except, in place of a circular antenna, they used a set of ultra-wideband (UWB) antennas that were arranged in a hemispherical shape to conform to the shape of the pendent breast [71]. The system's performance was evaluated based on experiments carried out on a phantom encapsulated in a shell mimicking skin. They were able to detect small tumor simulants, 4-6 mm in diameter, embedded in the phantoms by implementing two beamforming algorithms, multi-static microwave imaging (MAMI) and delay-and-sum (DAS). Later, the system was tested on highly heterogeneous phantoms and it was shown to give promising results [72]. Also, in a clinical case study, the system was tested on a real breast, and the obtained image was compared to its associated mammogram. The malignant tissue was

Lai and his group also employed UWB antennas to scan a number of heterogeneous and homogenous breast phantoms at Nanyan Technological University [74]. By rotating the phantoms 360 degrees, the antennas simulated a circular set of 360 elements. This configuration enabled them to detect a 4 mm tumor in all phantom types. But, detectability was not strong enough to delineate 2 mm inclusions. Lazero et al. published the results of what they named a "worst-case-study" utilizing a single UWB monopole antenna [75]. The phantom model was filled with water (presenting high losses), with a rod submerged inside it, and the antenna was located in air (presenting high reflection). Notwithstanding these challenging conditions, the embedded rod was successfully detected, implementing FDFD and Wiener-filter algorithm that took into account the effect of a simulated skin interface. Chun Yu and colleagues, in a different experimental layout, used an active microwave imaging method to retrieve images of a clay ball, a metallic ball, and a combined arrangement of both when they were submerged in water [76]. They utilized a dipole single transmitter and receiver to avert the mutual coupling that exists in an antenna array. The transmitting and receiving array were moved by an automatic positioner to enable a flexible, multi-view data acquisition in three dimensions over the surface of the object being measured. Applying an active, hybrid-inversion, reconstruction method, they were able to image one clay ball and the combined arrangement of two clay balls and one metallic ball.

correctly located in the microwave image when compared to the X-ray image [73].

Microwave imaging via space-time (MIST) beamforming has been experimentally implemented on phantoms, seeking to solve both 2D and 3D problem. On the simulation level, Bond et al. used a UWB antenna array to transmit the waves and then employed a beam former to reconstruct images of the backscattered signal energy [80]. In order to downsize the overshadowing impact of the skin-like layer of the MRI breast model they used, they utilized a data-adaptive algorithm to eliminate the dominant backscatter at the skin-breast interface. This approach is efficient in that it generates images only in areas of high background energy, i.e. malignant tissues. In the subsequent year, from the same group, Xi Lu et al. tested the practicality of MIST on 3D physical phantoms [81]. The results in 2D and 3D improved considerably compared to those obtained by simple focusing methods. In the 3D case, the contrast between malignant and normal tissue was observed to be 1.5:1 for a 4-mm tumor, which displayed the strong potential of MIST in disclosing small tumors.

At the end of this subsection, some of the recent works that are different in application or scope from the ones previously introduced can be reviewed to show how the practical aspect of microwave imaging has expanded in last few years. In 2014, Eleutério et al. presented a preliminary study to evaluate NRI in the axilla where sentinel nodes are typically found [82]. The significance of this study was to explore the role of microwave imaging in the estimation of metastasis initiation. Grzegorczyk et al. applied microwave imaging to monitor neoadjuvant chemotherapy, as an economic alternative to MRI and PET [83]. Using a different and compact method, researchers at the University of Manitoba examined the capacity of spintronic microwave sensors in biomedical imaging [84]. They demonstrated that these sensors, though only a fraction of microwave antennas in size, have the ability to detect spherical objects inside a homogenous medium with good resolution. Contrast-enhanced NRI has also become a topic of interest in the recent years. In one of the latest works published in 2016, Bucci et al. used magnetic nanoparticles as contrast agent, and showed their applicability by comparing the simulation and laboratory results [85].
