Free-Space Operating Microwave Imaging Device for Bone Lesion Detection: A Phantom Investigation

In this letter, a phantom validation of a low-complexity microwave imaging device operating in free space in the 1–6.5 GHz frequency band is presented. The device, initially constructed for breast cancer detection, measures the scattered signals in a multibistatic fashion and employs an imaging procedure based on the Huygens principle. Detection has been achieved in both bone fracture lesion and bone marrow lesion scenarios using the superimposition of five doublet transmitting positions, after applying the rotation subtraction artifact removal method. A resolution of 5 mm and a signal-to-clutter ratio (3.35 in linear scale) are achieved confirming the advantage of employing multiple transmitting positions on increased detection capability.


I. INTRODUCTION
A MONG the various imaging techniques, microwave imaging is accepted as an attractive technique for diagnostic applications due to its capabilities and safety to produce images of human organs [1]. Microwave imaging can be used for detection and monitoring of a variety of diseases since there is a contrast at microwave frequencies between the dielectric properties of healthy tissues and tissues with lesions [2]. Microwave imaging techniques are based on the scattering of electromagnetic waves. Specifically, a lesion immersed in a healthy tissue when illuminated by an electromagnetic field, transmitted by a dedicated antenna may generate a scattered field, which, if appropriately measured and processed, can be used to detect and locate the lesion itself.
Wide-ranging research has been performed to build and develop scanning systems using microwave imaging techniques suitable for medical applications. Many microwave imaging devices have been designed for breast lesion [3], [4] and brain stroke [5]- [7] detection. Recent  for bone imaging: for instance, Ruvio et al. [8] designed a microwave scanning system that consists of an antenna array immersed in a matching liquid. Bone lesion detection through microwave imaging has also been investigated via phantom measurements in an anechoic chamber, using two antennas in free space employing an imaging procedure based on the Huygens principle [9]. Recently, a portable microwave imaging device, operating in free space with two azimuthally rotating antennas has been constructed and used for breast cancer detection [10]. Specifically, the two antennas rotate around the breast to collect the signals in a multibistatic fashion. The purpose of this letter is to investigate and quantify, through phantom experiments, the capability of this imaging device to detect bone lesions. In this context, multilayered phantoms mimicking bone fractures or bone marrow lesions are realized using millimetric, cylindrically shaped inclusions to emulate lesions. An artifact removal procedure has been completed using a rotation subtraction method [9] comprising performing imaging after subtracting two measurements collected using two slightly displaced transmitting positions. Subsequently, a rigorous image quantification procedure has been implemented to assess the detection capability in two scenarios, i.e., bone fracture and bone marrow lesion. Finally, to evaluate detection capability in a more realistic scenario, the measurement was repeated by the cylindrically shaped inclusion with an inclusion having a high-eccentric elliptical cross section, i.e., flat-shaped. It is worthwhile to point out that the capability of Huygens principle-based microwave imaging to detect bone lesions was first demonstrated through phantom measurements inside an anechoic chamber [9]; here, instead we performed a phantom investigation using a microwave imaging device [named MammoWave, shown in Fig. 1(a)] [10].

A. Microwave Imaging Device Description
The microwave imaging device entails an aluminum cylindrical hub containing two antennas, one transmitting (tx) and one receiving (rx). The hub is internally covered by microwave absorbers, and is equipped with a hole and a cup, allowing the insertion of the object to be imaged. The antennas are installed at the same height, in free space, and can rotate around the azimuth to collect microwave signals from different angular positions [as shown in Fig. 1(b)]. More details of the device can be found in [10]. The tx and rx are connected to a two-port vector network analyzer (VNA) (S5065, Copper Mountain, Indianapolis, IN,   Fig. 1(c) illustrates the setup configuration. As Fig. 1(c) shows, in each doublet the two transmitting positions are displaced by 9 • . For each tx and rx position, the complex S 21 is collected from 1 to 6.5 GHz, with 5 MHz sampling. This frequency range can be considered appropriate for bone imaging [8], [9].

B. Phantom Descriptions
Two bone lesion phantoms comprising three layers have been fabricated. The three layers mimic: 1) the cortical bone (external layer); 2) the bone marrow layer (internal layer); and 3) a lesion. In the first phantom, which represents a bone fracture, the lesion has been placed between the external and internal layers, as shown in Fig. 2(a). In the second phantom [see Fig. 2 representing a bone marrow lesion, the lesion has been placed inside the internal layer. Dedicated liquids have been purchased from the ZMT Zurich MedTech Company [11] to mimic the different bone layers. The TLe11.5C.045 liquid oil has been employed to represent the external layer, which mimics the cortical bone layer. The internal layer, imitating the bone marrow layer, has been fabricated using Tle5C24 liquid oil. The lesion has been mimicked through a combination of 40% glycerol and 60% water [12]. The permittivity and conductivity values for TLe11.5C.045, Tle5C24, and the combination of 40% glycerol and 60% water are: ε r = 7 and σ = 0.3 S/m; ε r = 5 and σ = 0.2 S/m; and ε r = 60 and σ=2 S/m, respectively. It should be highlighted that the permittivity and conductivity values are calculated at 2 GHz and such dielectric properties can be considered representative of cortical bone, bone marrow, and a lesion constituted of blood [9].
Two cylindrically shaped plastic containers with diameters of 11 and 7 cm have been used to maintain the bone cortical and bone marrow equivalent materials, respectively. A cylindrically shaped tube of 4 mm diameter has been employed to contain an inclusion (see Fig. 2). The phantoms have been placed inside the cup of the microwave imaging device as shown in Fig. 1(b); in more details, the bone marrow lesion phantom has been positioned as shown in Fig. 3(a), whereas the bone fracture lesion phantom has been located as shown in Fig. 3(b). The cortical bone layer, bone marrow layer, and inclusion are presented in gray, light blue, and red, respectively. Additionally, two further scenarios have been considered, slightly modifying the bone marrow lesion phantom. First, the cylindrically shaped inclusion was replaced with an inclusion having a high-eccentric elliptical cross section, i.e., flat-shaped (major axis 6 mm, minor axis 2 mm). Next, the TLe11.5C.045 was substituted with a combination of 90% glycerol and 10% water, having ε r =16 and σ=1.1 S/m at 2 GHz [12]; such dielectric properties can be considered representative of a muscle and fat tissue [13].

C. Imaging Procedure
The measured complex S 21 has been processed through an imaging algorithm based on the Huygens principle [14] to generate the images. Assuming that rx can be rotatably moved to measure the received signal at the points rx np ≡ (a 0 , φ np ) ≡ − → ρ np displaced along a circular surface having radius a 0 , the received signals can be expressed as S 21np,m (a 0 , φ np ; tx m ; f ) with np = 1,2,...,80, m = 1,2,...,10, and f representing the frequency. The rotation subtraction strategy has been implemented for artifact removal by employing (1). This procedure has been performed through subtraction between transmitting position m and transmitting position m + 1, with m and m + 1 belonging to the same doublet where (ρ, φ) ≡ − → ρ is the observation point, Δs is the spatial sampling, and N P T is the number of receiving points. The component k 1 indicates the wavenumber, and G is the Green's function. The "reconstructed" internal field is indicated by the string "rcstr," whereas the string HP indicates that the Huygensbased procedure will be employed in (1). More details can be found in [14] and [15]. The intensity of the final images is given by the summation of different images corresponding to different transmitting doublets, each one obtained through the noncoherent summation of all frequency contributions. Therefore, the combined image of five transmitting position doublets has been generated through the following equation (N F being the number of frequency samples):

D. Imaging Quantification
To quantify the algorithm's detection capabilities, two metrics have been introduced and calculated: 1) resolution and 2) signal to clutter ratio (S/C). Specifically, the resolution is defined as a dimension of the region whose normalized intensity is above 0.5 [15]; S/C was defined as the ratio between maximum intensity evaluated in the region of the lesion divided by the maximum intensity outside the region of the lesion [16].  Table I for both bone fracture and bone marrow lesions. Fig. 6(a) and (b) shows the images obtained employing one single sample frequency, i.e., a single frequency contribution calculated for the central frequency (3.75 GHz), for bone fracture lesion and bone marrow lesion, respectively. Finally, Fig. 7(a) and (b) shows the images corresponding to the combination of five transmitting position doublets obtained for the bone marrow    lesion phantom when using a flat-shaped inclusion and after replacing the TLe11.5C.045 with a combination of 90% glycerol and 10% water, respectively. Both images are the combination of five transmitting position doublets.

IV. DISCUSSION
In order to investigate the ability of the proposed microwave imaging device to detect bone lesions, experiments have been performed using realistic phantoms with a very thin inclusion (diameter 4 mm), employing both individual transmitting positions doublets and their combination.
Rotation subtraction artifact removal has been employed to suppress artifacts, i.e., the image of the transmitter and the reflections of the layers. However, it may happen that even beyond artifact removal, residual clutter may mask the inclusion. Residual clutter is due to the imperfect cancelation of the transmitting antenna, inappropriate cancelation of the first layer reflection or can be due to multiple reflections occurring inside the phantom that cannot be canceled completely.
For the bone fracture lesion, detection is achieved using the second transmitting position doublet [ Fig. 4(b)], third doublet [ Fig. 4(c)], and fourth doublet [ Fig. 4(d)]. However, detection is not achieved using the first doublet [ Fig. 4(a)] and fifth doublet [ Fig. 4(e)], most likely due to residual clutter. Detection is also achieved using the combination of five doublets [ Fig. 4(f)]. For the bone marrow lesion case, as shown in Fig. 5, detection has been successfully achieved in all individual transmitting position doublets and also using the combination of the five doublets.
Concerning image quantification (see Table I), in bone fracture lesion the S/C varies up to 1.47 by employing the individual doublets, whereas in the bone marrow lesion scenario S/C varies from 1.53 to 2.58 using the individual doublets. These values are in good agreement with those from the anechoic chamber measurements [9], where a transmitting position doublet displaced at 5 • or 10 • was employed. Furthermore, Table I lists that, when using the combination of five doublets, S/C increases up to 2.26 in bone fracture lesion and 3.35 in bone marrow lesion. It follows that the combination of five doublets is beneficial in terms of S/C, a finding that is in agreement with previous studies [14]- [17].
The higher values of S/C for bone marrow lesion with respect to the bone fracture may be related to the existence of more residual clutter in the fracture lesion scenario, as the fracture lesion scenario is more asymmetric. This is confirmed from Fig. 6, which refers to one single frequency contribution.
Concerning the resolution, according to the collected data in Table I, using the combination of five doublets for both bone fracture and bone marrow lesion, the best resolution value equal to 5 mm was achieved. This is in excellent agreement with the optical resolution limit of λ 1,f max /4, where λ 1,f max represents the wavelength when considering a dielectric constant equal to the arithmetical average of the two layers calculated at the highest frequency of 6.5 GHz [15], [16]. Fig. 7(a) shows detection of the flat-shaped inclusion; interestingly, two spots appear in correspondence of the major axis extremities. This might suggests that lesions of 2 mm could be detected, even if, in the images, they might appear, larger in size. Fig. 7(b) shows detection also when a muscle and fat mimicking layer is employed as an external layer. Fig. 7 confirms that detection can be achieved in more realistic phantoms, even if an S/C lower than that of Fig. 5(f) can be (visually) appreciated. However, a limitation of the investigation presented in this letter is that phantoms having a cylindrically shaped external layer were always used. Next steps will be focused on the construction and use of anthropomorphic phantoms.

V. CONCLUSION
The capability of HP-based microwave imaging to detect bone lesions was first demonstrated through phantom measurements in an anechoic chamber [9]. Here, a phantom investigation using a microwave imaging device (based on HP) for bone lesions detection has been performed. Resolution of 5 mm and the S/C of 3.35 have been obtained by employing the combination of five transmitting position doublets in the imaging procedure and using a frequency band of 1-6.5 GHz. The microwave imaging device is safe (no X-rays), portable and it has low complexity since it employs only two rotating antennas operating in free space coupled through a VNA.