Access the full text.

Sign up today, get DeepDyve free for 14 days.

Applied Sciences
, Volume 8 (9) – Sep 6, 2018

/lp/multidisciplinary-digital-publishing-institute/a-single-simulation-platform-for-hybrid-photoacoustic-and-rf-acoustic-FVTILUis7Z

- Publisher
- Multidisciplinary Digital Publishing Institute
- Copyright
- © 1996-2019 MDPI (Basel, Switzerland) unless otherwise stated
- ISSN
- 2076-3417
- DOI
- 10.3390/app8091568
- Publisher site
- See Article on Publisher Site

applied sciences Article A Single Simulation Platform for Hybrid Photoacoustic and RF-Acoustic Computed Tomography 1 2,3, Christopher Fadden and Sri-Rajasekhar Kothapalli Department of Electrical Engineering, The Pennsylvania State University, University Park, PA 16802, USA; czf41@psu.edu Department of Biomedical Engineering, The Pennsylvania State University, University Park, PA 16802, USA Penn State Hershey Cancer Institute, The Pennsylvania State University, Hershey, PA 17033, USA * Correspondence: szk416@engr.psu.edu Received: 3 July 2018; Accepted: 29 August 2018; Published: 6 September 2018 Abstract: In recent years, multimodal thermoacoustic imaging has demonstrated superior imaging quality compared to other emerging modalities. It provides functional and molecular information, arising due to electromagnetic absorption contrast, at ultrasonic resolution using inexpensive and non-ionizing imaging methods. The development of optical- as well as radio frequency (RF)-induced thermoacoustic imaging systems would beneﬁt from reliable numerical simulations. To date, most numerical models use a combination of different software in order to model the hybrid thermoacoustic phenomenon. Here, we demonstrate the use of a single open source ﬁnite element software platform (ONELAB) for photo- and RF-acoustic computed tomography. The solutions of the optical diffusion equation, frequency domain Maxwell’s equations, and time-domain wave equation are used to solve the optical, electromagnetic, and acoustic propagation problems, respectively, in ONELAB. The results on a test homogeneous phantom and an approximate breast phantom conﬁrm that ONELAB is a very effective software for both photo- and RF-acoustic simulations, and invaluable for developing new reconstruction algorithms and hardware systems. Keywords: photoacoustic imaging; tomography; thermoacoustic; radio frequency 1. Introduction One of the goals of modern medical imaging is to simultaneously provide molecular, functional, and structural/anatomical information corresponding to various tissues. To achieve such comprehensive information, a combination of conventional imaging technologies such as X-ray computed tomography (CT) [1], positron emission tomography (PET) [2] and magnetic resonance imaging (MRI) [3] are used. CT provides anatomical contrast, PET provides molecular and metabolic information, and MRI provides both functional and anatomical contrasts. A PET–CT or PET–MRI combination is therefore widely used for simultaneously mapping molecular and anatomical contrasts [4–6]. All of these methods have signiﬁcant downsides: one of which is the ﬁnancial burden, which means that these methods are not ideal for routine imaging, and a second is the ionizing radiation used in CT and PET. As an alternative, ultrasound imaging that uses non-ionizing radiation is routinely used in several clinical applications for anatomical imaging [7]. However, it lacks the molecular or functional information necessary for detecting the early symptoms of disease. Alternative medical imaging modalities that provide anatomical, functional, and molecular information about the tissue, while being lower cost and not being as restrictive to patient movement, Appl. Sci. 2018, 8, 1568; doi:10.3390/app8091568 www.mdpi.com/journal/applsci Appl. Sci. 2018, 8, 1568 2 of 16 are needed. Photoacoustic computed tomography (PACT) and radio frequency (RF)-induced acoustic computed tomography (RACT), together known under the general term thermoacoustic computed tomography (TACT), match these objectives of lower cost functional/molecular imaging, using non-ionizing electromagnetic radiation [8–16]. While PACT maps optical absorption contrast using optical radiation induced acoustic wave detection, RACT maps tissue conductivity using RF induced acoustic wave detection. More importantly, since these hybrid (combining electromagnetic radiation and acoustic detection) imaging modalities share the same ultrasound detection platform, combined trimodality PACT–RACT–UCT (ultrasound computed tomography) systems have been realized for mapping functional, molecular, and anatomical contrasts [17]. The molecular absorption of electromagnetic energy causes thermal expansion in the tissue, which then leads to generation of acoustic waves. The acoustic waves propagate out of the tissue and are received by ultrasound transducers located at the boundary of the body. This data is then used to reconstruct thermoacoustic images displaying electromagnetic absorption contrast at ultrasonic spatial resolution. The imaging depth and spatial resolution in TACT is scalable with the frequency of excitation radiation and ultrasound transducer. The fact that the detected acoustic signal arises directly from speciﬁc molecules inside the tissue makes TACT a molecular/functional imaging technology. In PACT, the tissue chromophores—such as oxy-hemoglobin, deoxy-hemoglobin, melanin, and lipids—absorb light photons in the wavelength range from 400 nm to 1200 nm. By using different wavelengths, which takes advantage of the resonance peaks in the absorption spectrum of the imaged molecules, the distribution of different molecules inside the tissue can be mapped. In RACT, a radio frequency source in the frequency range from 434 MHz to 9 GHz is used for mapping the water distribution. The thermoacoustic effect in this frequency range is dependent on the conductivity distribution of the medium. The conductivity difference between water, tissue, and tumors can then give useful functional images. Accurate numerical modeling of TACT is paramount for the development of robust reconstruction algorithms to quantify the electromagnetic absorption properties of the tissue. In PACT, the forward optical simulation of total light ﬂuence (F), calculated inside the tissue medium, is usually achieved using either Monte Carlo simulations or the software package NIRFast, which solves the light diffusion equation. The ﬂuence distribution is then converted to an initial pressure rise, which is further propagated through the tissue medium and detected by the ultrasound transducers located on the boundary using acoustic simulation tools such as the K-wave toolbox. NIRFast [18] uses the ﬁnite element method (FEM), while K-wave uses a spectral-based ﬁnite difference method (FDM) to model the propagation of acoustic waves [19]. The goal of the reconstruction problem is then to recover the tissue properties (optical absorption in PACT and conductivity in RACT) given only the sensor data. Groups that use a single simulation platform for both the optical and acoustic propagation needed for the hybrid PACT/RACT technique are less common. Recently, there are studies that use the commercial software COMSOL to solve both propagation problems with a single software package [20,21]. As an alternative to the commercial software, the simulation system described in this paper uses the open source softwares Gmsh [22] and GetDP [23], often combined under the name ONELAB [24]. ONELAB is an FEM solver, which uses Gmsh for creating the FEM mesh, and GetDP for solving generic partial differential equations (PDEs) with the FEM method. Advantages of using Gmsh include its ability to create user deﬁned meshes, but also having standard interfaces with other commonly used mesh and computer-aided design (CAD) software such as STEP, IGES, and STL. Segmented DICOM images commonly used in MRI can then be converted to a mesh that Gmsh understands, creating realistic phantoms on which to test algorithms. The ability of GetDP to solve generic PDEs allows the user to implement algorithms to solve for optical and acoustic propagation, similar to the combination of NIRFast and K-wave. Both propagation methods, as well as any reconstruction methods, are implemented in the same mesh, with no loss of precision. Since there is less chance of numerical error, a single software platform represents a more accurate approximation of a real-world scenario. Besides PACT with optical sources, we demonstrate that the generality Appl. Sci. 2018, 8, 1568 3 of 16 of GetDP allows for sources in the RF regime of the electromagnetic spectrum to also be simulated for RACT. The use of the ONELAB software package allows accurate TACT modeling, in order to develop algorithms for functional imaging of the human body. Although there are experimental studies combining PACT and RACT techniques into one setup, perhaps our study is the ﬁrst to report a single TACT simulation platform for simulating both PACT and RACT. The rest of the paper is organized as follows: In Section 2, we describe methods and materials. Sections 3.1 and 3.2 show PACT and RACT results on a homogeneous phantom. Sections 3.3 and 3.4 show PACT and RACT results on an approximate breast phantom. Section 4 is a discussion of the results in Section 3, comparing errors in reconstruction between PACT and RACT, and between the two phantoms. Section 5 concludes the paper. 2. Materials and Methods There are three main phenomena that need to be modeled in thermoacoustics. The first is the propagation of the initial energy source, in this study near infrared optical or radio frequency electromagnetic waves, and calculation of the total fluence/intensity distribution inside the tissue medium. The second phenomenon is the acoustic propagation, after calculating an initial pressure from the intensity distribution. The third phenomenon is the reconstruction of the tissue parameters, which depend on the type of energy source used for tissue excitation. A ﬂowchart showing the similar steps between RACT and PACT is shown in Figure 1. This study is focused on using a time reversal [25,26] reconstruction method to recover the initial pressure, followed by a simple division to ﬁnd the tissue parameters. There are several alternative approaches for reconstructing the tissue parameters that can be integrated with our study in the future. Back-projection methods may be used when using speciﬁc geometries [27] and parametrix methods for more general geometries [28]. The back-projection methods are analytically exact, but only for geometries with a special symmetry, and with a constant speed of sound. Parametrix methods do not provide analytically exact methods, but instead give approximations with error bounds for general geometries, and can even be developed for regions with varying speeds of sound. A survey of several reconstruction methods is given in Reference [29]. Electric Field RF-Induced Recover Conductivity RF Recovered Intensity Initial Distribution Pressure Conductivity Simulation Pressure Rise (B ) (H ) (I ) 1 1 1 (C ) (D ) 1 1 Detect Acoustic Acoustic Finite Acoustic Forward Time-Reversal Element Mesh Pressure at Simulation Simulation (A) Boundary (E) (G) (F) Optical Optical Optical-Induced Optical Recover Absorption Fluence Initial Recovered Absorption Distribution Simulation Pressure Rise Pressure (I ) (B ) (C ) (D ) (H ) 2 2 2 2 Figure 1. The similarities between key steps of the photoacoustic computed tomography (PACT, blue) and radio frequency (RF)-induced acoustic computed tomography (RACT, red) workﬂows, with methods that are shared represented in yellow. Step A represents the ﬁnite element mesh in Section 2.5, used for both types of simulation. Step B is represented in Figure 9a, while step B is represented in Figure 7a. Step C uses the optical diffusion Equation (1). Step C uses the 2 2 1 electromagnetic wave Equation (5). The resulting pressure rise for the optical (D ) and RF (D ) 2 1 radiations are represented by Figures 7b and 9b, respectively. The initial pressure propagates out to the boundary (E) via the scalar wave Equation (7). The data received by acoustic sensors at the boundary (F) can then be used in reconstruction algorithms (G) to recover the initial pressure. In this paper, a time-reversal algorithm is applied for reconstruction (Section 2.4). The reconstructed pressure is then divided by the intensity (I , I ) to recover the recover the absorption or conductivity for the optical and 2 1 RF cases, shown in Figures 7c and 9c, respectively. Appl. Sci. 2018, 8, 1568 4 of 16 2.1. Optical Propagation for Photoacoustic Computed Tomography (PACT) A full description of optical propagation would use the radiative transport equation in its full generality. This equation can be modeled by the use of Monte Carlo methods, with signiﬁcant computational cost [30,31]. Since the the photoacoustic effect depends on the optical ﬂuence rate, and the wavelengths used correspond to the near infrared (NIR) regime, a diffusion approximation to the radiative transport equation is often used [32]. The diffusion equation for the ﬂuence rate is: r k(x)rF(x, l) + m (x)F(x) = q(x) x 2 W (1) F(x) = 0 x 2 ¶W. (2) In this equation, F is the optical ﬂuence rate, or intensity of the light. m is the absorption coefﬁcient at each point in space, while k = is the diffusion coefﬁcient, calculated using the absorption 3(m +m ) a s coefﬁcient and the reduced scattering coefﬁcient m . q is the source function, the initial laser pulse that irradiates the tissue medium. W is the tissue domain of interest, usually a subset of R , with n corresponding to the dimension of the problem (2 or 3). Equation (2) corresponds to the Dirichlet boundary condition used in this study. The amount of pressure generated by the optical energy is given av by a constant of proportionality, the Gruneisen parameter G = , where a is the volume thermal expansion coefﬁcient, v is the speed of sound, and C is the heat capacity at constant pressure [33]. s p Therefore, for a given ﬂuence rate F, the initial pressure is given by: p (x) = G(x)F(x)m (x). (3) 0 a While G will physically vary over space, this small deviation is ignored in this study, and the parameter is assumed constant, with a value of 0.1, which approximates standard tissue. 2.2. Radio Frequency Propagation for RF Acoustic Computed Tomography (RACT) Radio frequency radiation is modeled by Maxwell’s equations, which when modeled at a single frequency can be reduced to the following wave equation: rr E(x) w m(x)#(x)E(x) = q(x) x 2 W. (4) In this equation, E is the complex electric ﬁeld, w = 2p f is the frequency, and m and # are the permeability and permittivity of the medium, respectively. In a material with electric loss, the permittivity is complex valued, with # = # j , and s = 0 for a medium with no loss. real When the simulation is restricted to two dimensions (assuming a T M polarization), Equation (4) can be reduced to a scalar Helmholtz equation: 2 2 r E (x) + k(x) E (x) = q(x) x 2 W, (5) 3 3 where k = is the wavenumber, and can be complex in a tissue with energy loss. For boundary conditions, a Dirichlet boundary condition similar to Equation (2) is implemented. However, this would not prevent the electromagnetic waves from reﬂecting off the boundaries of the domain, causing numerical artifacts. Therefore, while there is a Dirichlet boundary condition, there is also a non-physical space occupied by a perfectly matched layer (PML) to absorb the outgoing waves and prevent reﬂection [34]. Once the electric ﬁeld is calculated, the initial pressure is given by three contributions [35]. The ﬁrst contribution is due to the conductivity of the medium, p = jEj dV. cond V 2 R R m m # # 0 r 2 0 r 2 The second due to the permittivity jEj dV, and the third the permeability j Hj dV. In V 2 V 2 practice, for the tissue media of interest in thermoacoustic tomography, the contribution to the pressure due to the conductivity dominates over the permittivity and permeability, and so these terms are Appl. Sci. 2018, 8, 1568 5 of 16 generally ignored. Therefore, similar to the optical absorption in Equation (3), the initial pressure due to the electric ﬁeld is given as: s(x) p (x) = G(x) jE(x)j . (6) 2.3. Thermoacoustic Equation Once an initial pressure is found, via either optical or radio frequency electromagnetic radiation, the pressure must then propagate to ultrasound transducers located outside the tissue boundary. Extremely short (nanosecond) pulses of energy are assumed to be irradiating the medium. Therefore the initial pressures are assumed to be delta functions, and only present for the initial conditions when formulating the equations. The pressure wave in photoacoustics is usually modelled using a standard scalar wave equation, with initial conditions: 2 2 p(x, t) = c (x)r p(x, t) x 2 W ¶t (7) p(x, 0) = f (x) p(x, 0) = 0, ¶t where, f (x) corresponds to the initial pressure found from the initial pulsed excitation, and p(x, t) is the pressure wave that propagates to the transducers. Similar to the electromagnetic wave equation, the pressure propagation also requires a perfectly matched layer (PML) to prevent reﬂections from the boundary of the numerical domain [36,37]. Therefore, the pressure is received at transducers along a curve (surface in 3D) g, which is not the boundary of the domain ¶W. 2.4. Time Reversal Reconstruction Algorithm The pressure received at the transducers can be represented as g(y, t), with y2g. The goal of the reconstruction problem is to reconstruct the tissue parameters, m in the optical excitation or s in the RF excitation case, given g(y, t). One way of recovering the parameters is to try and ﬁnd the initial pressure f (x) from the pressure measurements g(y, t), and then divide the pressure by the assumed known energy distribution to recover the parameters. One of the more common methods to recover the initial pressure is using the time-reversal technique. For the time-reversal to rigorously reconstruct f (x), requires Huygen’s principle to be valid. Unfortunately, Huygen’s principle does not hold when the speed of sound is not constant, and more problematically, it does not hold in two dimensions [38]. Therefore, in most domains of interest the time-reversal method recovers an approximation of the initial pressure f (x). As the name suggests, time reversal entails taking the received pressure g(y, t), and using it as a source on g, with time moving in reverse. If the forward simulation was run until a stoppage time T, then when simulated using the time reversal method p(x, T) f (x). There are several other methods that can be used to recover f (x) besides time-reversal, but there is no universally accepted reconstruction algorithm that works in all cases of interest. Most of these methods require pressure information at the sensors as functions of time. For this reason, the computationally expensive time-domain wave equation was used for solving the pressure wave equation instead of a Helmholtz equation similar to Equation (5). Since the initial pressure acts as a delta function, the received signal is inherently broadband. A time domain simulation is more efﬁcient for measuring broadband response, and provides a better representation of what would be measured during a physical experiment. 2.5. Phantom Geometry Two phantoms are used to demonstrate the effectiveness of the single simulation tool used in this study. The ﬁrst homogeneous phantom consists of a circular region of interest with radius 20 mm. Within this region, two objects are placed, a circle of radius 1 mm placed approximately 4 mm to the Appl. Sci. 2018, 8, 1568 6 of 16 left of center, and an ellipse with major axis 2 mm and minor axis 1 mm placed 4 mm to the right of center. The second phantom consists of an approximation to a human breast. The breast region of interest is represented by a circle of radius 50 mm. Glandular tissue rendered as a circle of radius 10 mm surrounds an elliptical tumor with major axis 7 mm and minor axis 3.5 mm. The tumor is located 32.5 mm deep from the right side of the phantom. The optical sources and acoustic detectors are placed around the region of interest in a continuous fashion. For creating the ﬁnite element meshes, a characteristic length of 2 mm was used for efﬁciency reasons, though near the absorbing objects, the size of the elements smoothly decreased in order to properly model the objects with sufﬁcient resolution. For the homogeneous phantom mesh, the total number of nodes in the mesh was 3635 with 7268 elements. The mesh for the breast phantom had 12,166 nodes and 24,330 elements. For the time domain simulation of the acoustic propagation, a time step of 30 nanoseconds was used in a Newmark numerical integration method. The homogeneous phantom ran for 700 time steps, while the breast phantom ran for 1700 time steps. A constant speed of sound of 1.5 mm/s was used for both the forward and time-reversed simulations. The time estimates for the breast phantom are: 0.6 s for the optical simulation, and 1.1 s for the RF simulation. The acoustic simulation took 18.1 s in the optical source case, and 17.3 s in the RF source case. The reconstruction using time reversal took 385.8 s and 408.0 s for the optical and RF source cases, respectively. The homogeneous phantom timings are: 0.16 s and 0.315 s for the optical and RF simulations; 2.1 s and 2.0 s for the forward acoustic simulation; and the time reversal took 38.4 s and 39.4 s for the optical and RF cases, respectively. The timings were done on a Dell Precision 5820 desktop PC, on a single thread. ONELAB has the functionality to run on multiple threads, as well as a graphical processing unit (GPU), but these options were not used in measuring these time estimates. 2.6. Phantom Parameters The absorption coefﬁcient m of the homogeneous phantom for the background was set to 0.001 mm . The circle and ellipse, which represent absorbers, have an absorption coefﬁcient of 0.425 mm ; the average absorption coefﬁcient of blood at 800 nm. As is common in human tissue, the background is assumed to have a higher reduced scattering coefﬁcient m than the absorbers. 1 0 The reduced scattering coefﬁcient for the background is set to 1 mm , while the absorbers m is set to zero for this phantom. In terms of electrical properties, the human body in general does not have a signiﬁcant magnetic response at radio frequencies, and so the relative permeability m is set to 1 for all objects. A background relative permittivity of # = 5 is similar to human tissue at 434 MHz. Unlike in the PACT case, objects such as tumors have a higher scattering coefﬁcient as well as absorption in the radio frequency regime. Therefore, the relative permittivity of the circle and ellipse were set to # = 25. The conductivity, which governs the amount of absorption, was set to 0.1 S/m for the background, approximating general tissue. The circle and ellipse are given values similar to that of a tumor, 10 S/m [39]. General breast tissue has an absorption coefﬁcient m = 0.005 mm , with a reduced scattering coefﬁcient of 1.52 mm when an 800 nm source is used. The optical properties of the glandular tissue and tumor depend on the assumed material composition of the tissues. The assumed amount of hemoglobin and percentage of water can have a dramatic effect on the properties at any given wavelength. In this study, the same composition as Reference [18] is used for the glandular and tumor tissue. The spectral characteristics of hemoglobin, deoxyhemoglobin, and water at 800 nm are taken from References [40,41]. The speciﬁc optical properties used for this breast phantom are given in Table 1. Appl. Sci. 2018, 8, 1568 7 of 16 Table 1. Optical properties (m the absorption coefﬁcient, and m the reduced scattering coefﬁcient) of different breast tissue at 800 nm. 1 0 1 Tissue m (mm ) m (mm ) Background 0.0005 1.5742 Glandular 0.0059 1.12 Tumor 0.0021 0.625 Human tissue is dispersive at radio frequencies, and so the electrical properties vary over frequency, though not as dramatically as the optical parameters. The properties for the generic and glandular tissue at 434 MHz were taken from the ITIS database provided by ETH-Zurich [42], which references a technical report compiled by the United States Air Force [43,44]. The tumor properties were extrapolated from data provided at 100 MHz in Reference [39] using a Cole–Cole dispersion model. The speciﬁc electromagnetic parameters are given in Table 2. The relative magnetic permeability is again set to unity, since the body does not exhibit strong magnetic response at 434 MHz. Table 2. Electrical properties (# the relative permittivity, and s the conductivity) of different breast tissue at 434 MHz. Tissue # s (S/m) Background 5.51 0.0353 Glandular 61.3 8.86 Tumor 25.25 13.03 3. Results The ability of the ONELAB software platform to simulate both PACT (optical source) and RACT (radio frequency source) with the same tool and on the same mesh, is demonstrated using the phantoms described in the methods section. Below we ﬁrst present PACT and RACT results for the homogeneous phantom embedded with two absorbers and then the results for the breast phantom with tumor. 3.1. Photoacoustic Computed Tomography (PACT) of the Homogeneous Phantom The optical excitation for the ﬁrst phantom uses a wavelength of 800 nm. This speciﬁc wavelength was chosen as it is often used in photoacoustics, due to the absorption spectrum of the absorbers of interest, such as hemoglobin. With the homogeneous phantom parameters, the simulated optical ﬂuence is given in Figure 2. The position of the absorbers is easily approximated by the nulls in the ﬂuence distribution. The approximate shapes can be identiﬁed, which can be used for more stable reconstruction of the material parameters. The 8 mm spacing is large enough to identify two distinct absorbers, with the ultimate resolution governed by the reconstruction of the ﬂuence that would be done in practice. The initial pressure induced by this ﬂuence distribution, as well as the original and reconstructed absorption coefﬁcient images are provided in Figure 3. The ﬂuence is not constant across the domain, and so the two absorbers induce slightly different initial pressures. The larger object is reconstructed with less error, since it is approximately constant over a larger area. Non-idealities in the reconstructed background pressure are suppressed by using the ﬂuence, and so the reconstructed objects are clearly separated from the background. 3.2. RF-Acoustic Computed Tomography (RACT) of the Homogeneous Phantom The RACT simulation used the same ﬁnite element mesh that was used for the photoacoustic simulation, utilizing a typically used radio frequency source of 434 MHz. For the parameters given in the methods section, the electric ﬁeld magnitude is provided in Figure 4. Appl. Sci. 2018, 8, 1568 8 of 16 Fluence Map (dB) 20 0 −2 −4 −6 −10 −8 −20 −20 −10 0 10 20 Cross-Range (mm) Figure 2. Total optical ﬂuence distribution inside the homogeneous phantom with two absorbers using an 800 nm source. The circular and elliptical absorbers, simulating hemoglobin with absorption coefﬁcient m = 0.425 mm at 800 nm, are located 4 mm to the left and right of the center. 1 0 1 0 The background has a m = 0.001 mm , and a reduced scattering coefﬁcient m = 1 mm . m = 0 a s s for the absorbers. −1 Initial Pressure (Pa) μ Distribution (mm ) 20 600 0.4 0.3 0 0 300 0.2 −10 −10 0.1 −20 −20 −20 −10 0 10 20 −20 −10 0 10 20 Cross-Range (mm) Cross-Range (mm) (a) Optical Absorption Distribution (b) Initial Pressure Rise −1 μ Distribution (mm ) Reconstructed Pressure (Pa) 0.3 300 0.2 0 0 0.1 −10 −10 −20 −20 −20 −10 0 10 20 −20 −10 0 10 20 Cross-Range (mm) Cross-Range (mm) (d) Reconstructed Pressure (c) Reconstructed Absorption Figure 3. Photoacoustic computed tomography (PACT) simulations using an 800 nm source on the homogeneous phantom with two light absorbing inclusions; (a) the true optical absorption distribution; (b) the initial pressure rise induced by the photoacoustic effect; (d) the reconstructed pressure, calculated using the time-reversal algorithm; and (c) the reconstructed absorption, obtained by dividing the reconstructed pressure with the ﬂuence. Unlike the PACT case, the electric ﬁeld alone does not provide any information on the location of the absorbers. Since the wavelength is much larger than the simulation domain, and there is small difference in conductivity between the absorbers and background, the electric ﬁeld intensity is approximately constant. The results of the RF acoustic simulation of this homogeneous phantom are shown in Figure 5. DownRange (mm) DownRange (mm) DownRange (mm) DownRange (mm) Down-Range (mm) Appl. Sci. 2018, 8, 1568 9 of 16 E-ﬁeld (dB) 20 0 −2 −5 · 10 −0.1 −10 −0.15 −20 −20 −10 0 10 20 Cross-Range (mm) Figure 4. Total electric ﬁeld intensity distribution inside the homogeneous phantom containing two absorbers, using a 434 MHz radio frequency (RF) source. The circular and elliptical absorbers, with conductivity s = 10 S/m, are located 4 mm to the left and right of the center, and have a dielectric constant # = 25. The background has s = 0.1 S/m, and # = 5. r r σ Distribution (S/m) Initial Pressure (Pa) 20 10 0 0 −10 −10 2 100 −20 −20 0 −20 −10 0 10 20 −20 −10 0 10 20 Cross-Range (mm) Cross-Range (mm) (b) Initial Pressure Rise (a) RF Conductivity Distribution σ Distribution (S/m) Reconstructed Pressure (Pa) 8 1 0 0 2 100 −10 −10 −20 −20 −20 −10 0 10 20 −20 −10 0 10 20 Cross-Range (mm) Cross-Range (mm) (d) Reconstructed Pressure (c) Reconstructed Conductivity Figure 5. Radio frequency (RF)-induced acoustic computed tomography (RACT) simulations using a 434 MHz RF source on the homogeneous phantom with two absorbers. (a) The true RF conductivity distribution; (b) the initial pressure rise induced by the thermoacoustic effect; (d) the reconstructed pressure, calculated using the time-reversal algorithm; and (c) the reconstructed conductivity, obtained by dividing the reconstructed pressure by the electric ﬁeld intensity. 3.3. Photoacoustic Computed Tomography (PACT) of the Breast Phantom Figure 6 shows the ﬂuence distribution generated by ONELAB for the breast phantom described in the methods section. A rough estimate of the tumor location can be predicted from the ﬂuence map, but no shape information can be obtained from the ﬂuence distribution alone. The photoacoustic simulation is able to identify the shape of the tumor, as well as the difference between cancerous and DownRange (mm) Down-Range (mm) DownRange (mm) DownRange (mm) DownRange (mm) Appl. Sci. 2018, 8, 1568 10 of 16 glandular tissue. Results in Figure 7 show the initial pressure rise and the resulting reconstructed absorption distribution. Fluence Map (dB) −2 −4 −20 −6 −40 −8 −40 −20 0 20 40 Cross-Range (mm) Figure 6. Total optical ﬂuence distribution using an 800 nm wavelength source on the breast phantom. The background absorption coefﬁcient is m = 0.0005 mm , with a reduced scattering coefﬁcient 0 1 0 1 m = 1.5742. Glandular tissue (m = 0.0059 mm , m = 1.12 mm ) surrounds a tumor (m = 0.0021, a a s s 0 1 m = 0.625 mm ) located 32.5 mm deep from the right side of the phantom. −1 μ Distribution (mm ) Initial Pressure (Pa) −3 · 10 40 40 20 20 - - −20 −20 2 50 −40 −40 1 −40 −20 0 20 40 −40 −20 0 20 40 Cross-Range (mm) Cross-Range (mm) (a) Optical Absorption Distribution (b) Initial Pressure Rise −1 Reconstructed Pressure (Pa) μ a Distribution (mm ) −3 · 10 40 40 0 0 −20 −20 −40 0 −40 −40 −20 0 20 40 −40 −20 0 20 40 Cross-Range (mm) Cross-Range (mm) (c) Reconstructed Absorption (d) Reconstructed Pressure Figure 7. Photoacoustic computed tomography (PACT) simulations of the breast phantom at 800 nm wavelength; (a) the true optical absorption; (b) the initial pressure rise induced by the photoacoustic effect; 1 1 (d) the reconstructed pressure, calculated using the time reversal algorithm; and (c) the reconstructed absorption, obtained by dividing the reconstructed pressure with the ﬂuence. The elliptical region similar to the background is the tumor, surrounded by the glandular tissue. 3.4. RF-Induced Acoustic Computed Tomography (RACT) of the Breast Phantom The radio frequency source for the breast phantom, similar to the homogeneous phantom, operates at 434 MHz. The ﬁeld distribution is provided in Figure 8. The forward simulation of the electric ﬁeld is able to directly detect the tumor and an estimation of its location. The conductivity has a much more signiﬁcant effect on the wave, since the wavelength 1 1 DownRange (mm) DownRange (mm) DownRange (mm) DownRange (mm) DownRange (mm) Appl. Sci. 2018, 8, 1568 11 of 16 is larger than the region of interest. In the large wavelength regime, scattering due to dielectric contrast does not distort the electromagnetic wave as much as the substantial conductivity. While the approximate location of the tumor can be inferred from the electric ﬁeld intensity, the speciﬁc shape and glandular tissue identiﬁcation requires further processing, such as the RF acoustic simulation. Figure 9 shows the initial pressure and the resulting reconstructed pressure and conductivity maps of the breast phantom. E-ﬁeld (dB) −1 −2 −3 −20 −4 −40 −40 −20 0 20 40 Cross-Range (mm) Figure 8. Total electric ﬁeld intensity distribution of a 434 MHz radio frequency source inside the breast phantom. The background has a conductivity s = 0.0353 S/m and dielectric constant # = 5.51. Glandular tissue (s = 8.86 S/m, # = 61.3) surrounds a tumor (s = 13.03 S/m, # = 25.25) located r r 32.5 mm deep from the right side of the phantom. σ Distribution (S/m) Initial Pressure (Pa) 40 40 20 20 0 0 - - 40 −20 4 −20 −40 −40 −40 −20 0 20 40 −40 −20 0 20 40 Cross-Range (mm) Cross-Range (mm) (a) RF Conductivity Distribution (b) Initial Pressure Rise σ Distribution (S/m) Reconstructed Pressure (Pa) 40 1 40 20 20 60 0 0 - - −20 −20 −40 −40 −40 −20 0 20 40 −40 −20 0 20 40 Cross-Range (mm) Cross-Range (mm) (c) Reconstructed RF Conductivity (d) Reconstructed Pressure Figure 9. RF-induced acoustic computed tomography (RACT) simulations of the breast phantom using a 434 MHz RF source; (a) the true RF conductivity; (b) the initial pressure rise induced by the thermoacoustic effect; (d) the reconstructed pressure, calculated using the time-reversal algorithm; 1 1 and (c) the reconstructed absorption, found by dividing the reconstructed pressure by the electric ﬁeld intensity. The elliptical region with signiﬁcantly larger conductivity is the tumor, surrounded by the glandular tissue. 1 1 DownRange (mm) DownRange (mm) DownRange (mm) DownRange (mm) DownRange (mm) Appl. Sci. 2018, 8, 1568 12 of 16 4. Discussion 4.1. Merits of ONELAB for Thermoacoustic Imaging Using an open source platform, ONELAB, allowed a single interface to simulate hybrid thermoacoustic imaging, with the choice of optical, radio-frequency, or other sources to induce a pressure wave response. The components of ONELAB Gmsh (mesh) and GetDP (FEM solver) were applied for this purpose. Since GetDP only solves the user deﬁned ﬁnite element equations, ﬁne control over every aspect of the thermoacoustic simulation was possible. This is especially important when performing reconstruction, since, for instance, time-reversal can have drastically different performance depending on the amount of time steps and the duration per time step. The beneﬁts of Gmsh include its standard mesh creation, with the ability to automatically interface with other standard mesh formats. This allows the user to write a wrapper around any given ﬁnite element mesh. Besides simulating expected results, this platform is also ideal for post-processing the results. All of the data from the simulated PACT and RACT experiments was in the same format, used the same ﬁnite element mesh, without the need for interpolation or data transformation. Using the data in such a complementary fashion demonstrated the beneﬁts of ONELAB over more established simulation tools used for a single imaging modality. 4.2. Simulation Workﬂow for both PACT and RACT We ﬁrst tested our algorithms on a simple phantom that consisted of two absorbing regions of different shapes inside a homogeneous background. The second phantom simulated a human breast with a tumor. The same phantom mesh was used to deﬁne both optical and RF-acoustic properties of the tissue. The optical diffusion equation was solved to obtain the optical ﬂuence maps of the phantoms (Figures 2 and 6), while the solution to Maxwell’s equations mapped the electric ﬁeld distribution for these phantoms (Figures 4 and 8). Subsequently, maps of the initial pressure rises, due to electromagnetic absorption and the thermoacoustic effect, were generated using the optical ﬂuence maps in PACT and electric ﬁeld intensity in RACT. The initial pressure distributions of both PACT and RACT were then propagated using the same time-domain equations. Respective maps of the reconstructed pressure were generated using a time-reversal algorithm. Optical absorption and conductivity of the tissue phantoms were recovered by dividing the related reconstruction pressures by the optical ﬂuence and electric ﬁeld intensity distributions, respectively (Figures 3, 5, 7 and 9). 4.3. Analysis of PACT and RACT Results Overall, our simulation results on two different tissue phantoms have shown that ONELAB can effectively simulate photoacoustic computed tomography (PACT) as well as RF-induced acoustic computed tomography (RACT). The optical ﬂuence distribution (Figure 2) and electric ﬁeld intensity (Figure 4) for the homogeneous phantom were signiﬁcantly different, though the PACT/RACT reconstruction (Figures 3 and 5) is of similar accuracy. The relative error of the maximum absorption coefﬁcient was 1.25% for the PACT case, and 8.70% in the RACT case. PACT for this phantom had a much lower relative error, since the contrast between absorber was much larger than the same difference in the RACT case. Imperfect reconstruction of the absorbers is due to error in reconstructing the initial pressure, instead of differences in the ﬁeld intensity distribution. The relative shape of the absorbers, using either PACT or RACT, is easily identiﬁed from the reconstruction, with further processing only necessary for very precise characterizations. The successfully simulated results on a generic phantom gave us conﬁdence to further validate our algorithms on a real tissue phantom mimicking the optical and RF properties of a human breast. When the approximate optical and RF parameters corresponding to breast tissue are used (Figures 6 and 8), the optical and electric ﬁeld intensities resemble each other, and have similar accuracy to the homogeneous phantom. The relative error of the maximum absorption coefﬁcient was 23.2% for PACT and 6.92% for RACT. In the realistic tissue, RACT had more contrast, leading to less error in Appl. Sci. 2018, 8, 1568 13 of 16 the reconstruction. The difference in ﬁeld intensity between the homogeneous and breast phantom did not have a signiﬁcant effect on the reconstruction accuracy. Both PACT and RACT are able to approximately reconstruct the tumor and the surrounding glandular tissue with no further processing. Even though the tumor had properties similar to the background, the PACT reconstruction was still able to identify the tumor surrounded by the glandular tissue. In RACT, the tumor is readily identiﬁed as being signiﬁcantly different from the background. In summary, our work demonstrated that ONELAB is a viable simulation platform for use in PACT and RACT, and is well suited for experiments that exploit both modalities. Reconstructed images of the phantoms provided both qualitative and quantitative information about the tissue optical and conductivity properties, including the size, shape, and location of the target regions. This work laid a foundation for future studies to develop and validate more robust multimodality reconstruction algorithms that will help improve quantitative accuracy. 5. Conclusions This study has demonstrated the use of the tools Gmsh and GetDP, known together as ONELAB, as a single simulation platform for modeling both optical, as well as RF-induced, thermoacoustic computed tomography, i.e. PACT and RACT, respectively. To achieve PACT and RACT results, the propagation of optical, radio frequency, and acoustic waves were effectively modeled using solutions of the optical diffusion equation, Maxwell’s equations, and time-domain wave equations. We validated our PACT and RACT algorithms using two types of tissue mimicking phantoms: a homogeneous phantom consisting of two absorbing targets and a breast phantom consisting of a tumor, with pre-deﬁned optical and RF properties. Our results demonstrated that the optical and RF absorption properties of the respective tissue phantoms were accurately reconstructed using the proposed dual-modality computed tomography simulations in ONELAB. The use of the ONELAB software package allows for accurate multimodal thermoacoustic modeling, in order to develop and validate more robust algorithms for functional imaging of the human body. Author Contributions: C.F. designed the simulations, analyzed the results, and wrote the paper. S.R.K. assisted in both the analysis of the results and writing of the paper. Funding: The authors would like to thank the NIH for funding this work under NIH-NIBIB 4R00EB017729 (Sri-Rajasekhar Kothapalli) and PennState Hershey Center Institute Cancer Startup Funds (Sri-Rajasekhar Kothapalli). Conﬂicts of Interest: The authors declare no conﬂict of interest. Abbreviations The following abbreviations are used in this manuscript: ONELAB Open Numerical Engineering Laboratory GetDP General Environment for the Treatment of Discrete Problems RF radio frequency CT computed tomography PET positron emission tomography MRI magnetic resonance imaging UCT ultrasound computed tomography PACT photoacoustic computed tomography RACT radio frequency acoustic computed tomography TACT thermoacoustic computed tomography FEM ﬁnite element method FDM ﬁnite difference method PDE partial differential equation CAD computer-aided design Appl. Sci. 2018, 8, 1568 14 of 16 IGES initial graphics exchange speciﬁcation STL stereolithography DICOM Digital Imaging and Communications in Medicine NIR near infrared TM transverse magnetic PML perfectly matched layer References 1. Glick, S.J. Breast CT. Annu. Rev. Biomed. Eng. 2007, 9, 501–526. [CrossRef] [PubMed] 2. Vercher-Conejero, J.L.; Pelegrí-Martinez, L.; Lopez-Aznar, D.; Cózar-Santiago, M.D.P. Positron Emission Tomography in Breast Cancer. Diagnostics 2015, 5, 61–83. [CrossRef] [PubMed] 3. Lehman, C.D.; Schnall, M.D. Imaging in breast cancer: Magnetic resonance imaging. Breast Cancer Res. 2005, 7, 215–219. [CrossRef] [PubMed] 4. Boone, J.M.; Yang, K.; Burkett, G.W.; Packard, N.J.; Huang, S.Y.; Bowen, S.; Badawi, R.D.; Lindfors, K.K. An X-ray Computed Tomography/Positron Emission Tomography System Designed Speciﬁcally for Breast Imaging. Technol. Cancer Res. Treat. 2010, 9, 29–44. [CrossRef] [PubMed] 5. Martí-Bonmatí, L.; Sopena, R.; Bartumeus, P.; Sopena, P. Multimodality imaging techniques. Contrast Media Mol. Imaging 2010, 5, 180–189. [CrossRef] [PubMed] 6. Torigian, D.A.; Zaidi, H.; Kwee, T.C.; Saboury, B.; Udupa, J.K.; Cho, Z.H.; Alavi, A. PET/MR Imaging: Technical Aspects and Potential Clinical Applications. Radiology 2013, 267, 26–44. [CrossRef] [PubMed] 7. Sehgal, C.M.; Weinstein, S.P.; Arger, P.H.; Conant, E.F. A review of breast ultrasound. J. Mammary Gland Biol. Neoplasia 2006, 11, 113–123. [CrossRef] [PubMed] 8. Ku, G.; Wang, L.V. Scanning thermoacoustic tomography in biological tissue. Med. Phys. 2000, 27, 1195–1202. [CrossRef] [PubMed] 9. Xu, M.; Wang, L.V. RF-induced thermoacoustic tomography. In Proceedings of the Second Joint 24th Annual Conference and the Annual Fall Meeting of the Biomedical Engineering Society Engineering in Medicine and Biology, Houston, TX, USA, 23–26 October 2002; Volume 2, pp. 1211–1212. [CrossRef] 10. Wang, X.; Pang, Y.; Ku, G.; Xie, X.; Stoica, G.; Wang, L.V. Noninvasive laser-induced photoacoustic tomography for structural and functional in vivo imaging of the brain. Nat. Biotechnol. 2003, 21, 803–806. [CrossRef] [PubMed] 11. Yang, X.; Wang, L.V. Monkey brain cortex imaging by photoacoustic tomography. J. Biomed. Opt. 2008, 13, 044009. [CrossRef] [PubMed] 12. Akhouayri, H.; Bergounioux, M.; Da, S.A.; Elbau, P.; Litman, A.; Mindrinos, L. Quantitative thermoacoustic tomography with microwaves sources. J. Inverse Ill-Posed Probl. 2016, 25, 703–717. [CrossRef] 13. Pramanik, M.; Ku, G.; Li, C.; Wang, L.V. Design and evaluation of a novel breast cancer detection system combining both thermoacoustic (TA) and photoacoustic (PA) tomography. Med. Phys. 2008, 35, 2218–2223. [CrossRef] [PubMed] 14. Razansky, D.; Kellnberger, S.; Ntziachristos, V. Near-ﬁeld radiofrequency thermoacoustic tomography with impulse excitation. Med. Phys. 2010, 37, 4602–4607. [CrossRef] [PubMed] 15. Heijblom, M.; Piras, D.; Brinkhuis, M.; van Hespen, J.C.G.; van den Engh, F.M.; van der Schaaf, M.; Klaase, J.M.; van Leeuwen, T.G.; Steenbergen, W.; Manohar, S. Photoacoustic image patterns of breast carcinoma and comparisons with Magnetic Resonance Imaging and vascular stained histopathology. Sci. Rep. 2015, 5, 11778. [CrossRef] [PubMed] 16. Emerson, J.F.; Chang, D.B.; McNaughton, S.; Jeong, J.S.; Shung, K.K.; Cerwin, S.A. Electromagnetic Acoustic Imaging. IEEE Trans. Ultrason. Ferroelectr. Freq. Control 2013, 60, 364–372. [CrossRef] [PubMed] 17. Cui, H.; Yang, X. In vivo imaging and treatment of solid tumor using integrated photoacoustic imaging and high intensity focused ultrasound system. Med. Phys. 2010, 37, 4777–4781. [CrossRef] [PubMed] 18. Dehghani, H.; Eames, M.E.; Yalavarthy, P.K.; Davis, S.C.; Srinivasan, S.; Carpenter, C.M.; Pogue, B.W.; Paulsen, K.D. Near infrared optical tomography using NIRFAST: Algorithm for numerical model and image reconstruction. Commun. Numer. Methods Eng. 2008, 25, 711–732. [CrossRef] [PubMed] 19. Treeby, B.E.; Cox, B.T. k-Wave: MATLAB toolbox for the simulation and reconstruction of photoacoustic wave ﬁelds. J. Biomed. Opt. 2010, 15, 021314. [CrossRef] [PubMed] Appl. Sci. 2018, 8, 1568 15 of 16 20. Wang, Z.; Ha, S.; Kim, K. Evaluation of ﬁnite element based simulation model of photoacoustics in biological tissues. In Medical Imaging 2012: Ultrasonic Imaging, Tomography, and Therapy; International Society for Optics and Photonics: San Diego, CA, USA, 2012; p. 83201L. [CrossRef] 21. Sowmiya, C.; Thittai, A.K. Simulation of photoacoustic tomography (PAT) system in COMSOL and comparison of two popular reconstruction techniques. In Medical Imaging 2017: Biomedical Applications in Molecular, Structural, and Functional Imaging; International Society for Optics and Photonics: Orlando, FL, USA, 2017; Volume 10137, p. 101371O. [CrossRef] 22. Geuzaine, C.; Remacle, J.F. Gmsh: A 3-D ﬁnite element mesh generator with built-in pre- and post-processing facilities. Int. J. Numer. Methods Eng. 2009, 79, 1309–1331. [CrossRef] 23. Dular, P.; Geuzaine, C.; Henrotte, F.; Legros, W. A general environment for the treatment of discrete problems and its application to the ﬁnite element method. IEEE Trans. Magn. 1998, 34, 3395–3398. [CrossRef] 24. Geuzaine, C.; Henrotte, F.; Remacle, J.F.; Marchandise, E.; Sabariego, R. ONELAB: Open Numerical Engineering LABoratory. In Proceedings of the 11e Colloque National en Calcul des Structures, Giens, France, 13–17 May 2013. 25. Treeby, B.E.; Zhang, E.Z.; Cox, B.T. Photoacoustic tomography in absorbing acoustic media using time reversal. Inverse Probl. 2010, 26, 115003. [CrossRef] 26. Cox, B.T.; Treeby, B.E. Artifact Trapping During Time Reversal Photoacoustic Imaging for Acoustically Heterogeneous Media. IEEE Trans. Med. Imaging 2010, 29, 387–396. [CrossRef] [PubMed] 27. Xu, M.; Wang, L.V. Universal back-projection algorithm for photoacoustic computed tomography. Phys. Rev. E 2005, 71, 016706. [CrossRef] [PubMed] 28. Popov, D.A.; Sushko, D.V. Image Restoration in Optical-Acoustic Tomography. Probl. Inf. Trans. 2004, 40, 254–278. [CrossRef] 29. Kuchment, P.; Kunyansky, L. Mathematics of thermoacoustic tomography. Eur. J. Appl. Math. 2008, 19, 191–224. [CrossRef] 30. Wang, L.; Jacques, S.L.; Zheng, L. MCML—Monte Carlo modeling of light transport in multi-layered tissues. Comput. Methods Programs Biomed. 1995, 47, 131–146. [CrossRef] 31. Kim, C.; Li, Y.; Wang, L.V. The study of quantitative optical absorption imaging by using Monte Carlo simulation of combined photoacoustic tomography and ultrasound-modulated optical tomography. In Photons Plus Ultrasound: Imaging and Sensing 2012; International Society for Optics and Photonics: San Francisco, CA, USA, 2012; Volume 8223, p. 82232C. [CrossRef] 32. Ammari, H. (Ed.) Mathematical Modeling in Biomedical Imaging II: Optical, Ultrasound, and Opto-Acoustic Tomographies; Mathematical Biosciences Subseries; Springer: Berlin/Heidelberg, Germay, 2012. 33. Wang, L.V. (Ed.) Photoacoustic Imaging and Spectroscopy, 1st ed.; CRC Press: Boca Raton, FL, USA, 2009. 34. Berenger, J.P. A perfectly matched layer for the absorption of electromagnetic waves. J. Comput. Phys. 1994, 114, 185–200. [CrossRef] 35. Kellnberger, S.; Hajiaboli, A.; Razansky, D.; Ntziachristos, V. Near-ﬁeld thermoacoustic tomography of small animals. Phys. Med. Biol. 2011, 56, 3433–3444. [CrossRef] [PubMed] 36. Roden, J.A.; Gedney, S.D. Convolution PML (CPML): An efﬁcient FDTD implementation of the CFS–PML for arbitrary media. Microw. Opt. Technol. Lett. 2000, 27, 334–339. [CrossRef] 37. Kaltenbacher, B.; Kaltenbacher, M.; Sim, I. A modiﬁed and stable version of a perfectly matched layer technique for the 3-d second order wave equation in time domain with an application to aeroacoustics. J. Comput. Phys. 2013, 235, 407–422. [CrossRef] [PubMed] 38. Evans, L.C. Partial Differential Equations: Second Edition, 2nd ed.; American Mathematical Society: Providence, RI, USA, 2010. 39. Surowiec, A.J.; Stuchly, S.S.; Barr, J.R.; Swarup, A. Dielectric properties of breast carcinoma and the surrounding tissues. IEEE Trans. Biomed. Eng. 1988, 35, 257–263. [CrossRef] [PubMed] 40. Jacques, S.L. Optical properties of biological tissues: A review. Phys. Med. Biol. 2013, 58, R37. [CrossRef] [PubMed] 41. Prahl, Scott. Assorted Spectra. Available online: https://omlc.org/spectra/index.html (accessed on 28 June 2018). 42. Lloyd, Bryn A. Dielectric Properties. Available online: https://itis.swiss/virtual-population/tissue- properties/database/dielectric-properties/ (accessed on 19 June 2018). Appl. Sci. 2018, 8, 1568 16 of 16 43. Gabriel, C. Compilation of the Dielectric Properties of Body Tissues at RF and Microwave Frequencies; Air Force Materiel Command, Brooks Air Force Base: San Antonio, TX, USA, 1996; p. 272. 44. Gabriel, C.; Gabriel, S.; Corthout, E. The dielectric properties of biological tissues: I. Literature survey. Phys. Med. Biol. 1996, 41, 2231. [CrossRef] [PubMed] © 2018 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).

Applied Sciences – Multidisciplinary Digital Publishing Institute

**Published: ** Sep 6, 2018

Loading...

You can share this free article with as many people as you like with the url below! We hope you enjoy this feature!

Read and print from thousands of top scholarly journals.

System error. Please try again!

Already have an account? Log in

Bookmark this article. You can see your Bookmarks on your DeepDyve Library.

To save an article, **log in** first, or **sign up** for a DeepDyve account if you don’t already have one.

Copy and paste the desired citation format or use the link below to download a file formatted for EndNote

Access the full text.

Sign up today, get DeepDyve free for 14 days.

All DeepDyve websites use cookies to improve your online experience. They were placed on your computer when you launched this website. You can change your cookie settings through your browser.