ICCSCE'14_Compressed Sensing Based Nearfield Electromagnetic Imaging

Compressed Sensing Based Nearfield Electromagnetic Imaging

Muhammad Naveed Tabassum, Ibrahim Elshafiey, Mubashir Alam Department of Electrical Engineering

King Saud University Riyadh, Kingdom of Saudi Arabia

[email protected], [email protected], [email protected]

AbstractThis paper proposes a novel method of nearfield electromagnetic imaging using compressed sensing technique. Orthogonal matching pursuit (OMP) reconstruction algorithm is implemented for reconstruction of the target space. A dictionary is tested considering head imaging of single and multiple brain tumor targets. The received scattered time-domain signals are captured using spatial compressed sensing and later interpolated for full target space. These signals are also processed for temporal compressed sensing using background subtraction. Simulation of the forward problem it is conducted using CST Microwave Studio using frequency range of 300-3000 megahertz. The quality of reconstructed images reveals the potential of the proposed method.

Index TermsNearfield imaging, electromagnetic imaging, brain tumor imaging, temporal compressed sensing, spatial compressed sensing.

I. INTRODUCTION Electromagnetic imaging (EMI) using RF and microwave

ranges is an attractive research topic in various disciplines including civilian, military, industrial, and biomedical fields. Application examples of EMI comprise non-destructive evaluation, remote sensing, and biomedical diagnosis and imaging [1-4]. For example, by employing simple and inexpensive systems, and using non-ionizing fields, EMI provide more attractive tools compared to other biomedical imaging modalities such as x-ray, CAT scan or MRI. The contrast in dielectric properties for cancerous and healthy body tissues have been used for tumor localization in applications such as breast imaging [2-4]. However, hitherto EMI of human head has proven to be more challenging [5]. The higher conductivity values for the brain tissues compared to that for breast tissues is also a reason for limited efforts in building a complete system for head imaging.

This work aims at enhancing biomedical head imaging for identifying brain tumor using compressed sensing based nearfield EMI. Therefore, the key objectives of head imaging here are to detect the presence and location of the brain tumor using both temporal and spatial compressed sensing using wideband excitation signals. Consequently, the imaging system consists of an applicator antenna array, a data collection system, and a post processing system that is used to analyze and invert the collected data to reconstruct required images.

Efficient nearfield RF and microwave imaging requires exorbitant data collection. Indeed, data collection for different wideband frequencies results in large amounts of scattered data that is to be collected using a large number of transmitter and receiver locations. This complex process is a major problem for the wideband nearfield RF and microwave imaging. Moreover, the interaction between the antenna array elements and biological tissues in nearfield makes the investigation more complex, since farfield approximations will be invalid.

Furthermore, large attenuation is encountered due to the biological tissues, which results in highly weak received scattered signals at the sensors. Additionally, the difference in distance from sensors for different possible tumor locations is also extremely small, and this results in highly correlated acquired signals. Advanced preprocessing analysis is necessary to extract useful information from these weak and highly correlated signals.

In this paper, compressed sensing (CS) is implemented in combination with background and reflection subtraction to overcome discussed challenges. Compressed sensing is an attractive emerging technique for data collection. The CS algorithm allows the accurate recovery of signals and images, and other data from what appear to be highly sub-Nyquist-rate samples [6, 7]. The CS makes use of signal sparsity in spectrum as described in [8]. Here temporal CS analysis is performed along with spatial interpolation. The adopted framework achieves reconstruction from fewer temporal samples (as compared to reconstructed image size), considering various possible tumor locations. CS technique provides thus considerable savings in required data acquisition time.

This paper is organized as follows. The detection of brain tumor by forward problem formulation and capturing of received signals using human head and antenna array is discussed in Section II. Dictionary creation for target space, which will be used for compressed sensing (CS) based reconstruction is illustrated in Section III. The CS based inverse problem solution and the inversion results are presented in Section IV and V respectively. Lastly, the paper is concluded in Section VI.

2014 IEEE International Conference on Control System, Computing and Engineering, 28 - 30 November 2014, Penang, Malaysia

978-1-4799-5686-9/14/$31.00 2014 IEEE 571

II. DETECTION OF BRAIN TUMAdvanced computational and electr

simulation is implemented to formulate thefor the detection of brain tumor as described is performed using CST Microwave Studiwhich provides solvers in time domain and of 3D EM simulation [9].

A. Human Head Model with Antenna Array A simple human head phantom model is

the computational analysis. The model concylindrical structures, where the inner cyaverage brain tissue with a radius of 85 mouter cylinder represents the skull and hamillimeter [10], as shown in Fig. 1. The diused in the computations for human head molisted in Table I, which are assigned accordreported in the literature for frequency of 911].

Horn antenna is used for the electromagnetic (EM) signals. The structurgiven in [12]. The operational range of wmicrowave frequencies is selected from 300This wideband frequency range providebetween penetration for the EM signals andthe reconstructed images. The applicator anteof four elements around the human head, asFigure 2 shows a typical Gaussian input signrequired frequency band, and is used for elem

Fig. 1. Human head model with tumor and antenna

MOR romagnetic (EM) e forward problem next. The analysis o MWS package, frequency domain

s implemented for ntains two layered ylinder represents

millimeter, and the as a radius of 90 ielectric properties odel and tumor are ding to the values 15 megahertz [10,

transmission of re of the horn is wideband RF and 0-3000 megahertz. es a compromise d the resolution of enna array consists s shown in Fig. 1. nal that covers the

ment excitation.

a array (top view).

TABLE I. DIELECTRIC PROPERTTUM

Tissue Type DielectrConstan

r Average Brain

Tissue 38.837

Skull 16.598

Tumor 63.259

Fig. 2. Input excitation signal

B. Raw Received Time-domain The scattered time-doma

recorded, at each element of tsignals can roughly be identifiein Fig. 3: the coupled signalantenna array, the reflected sigreflections for the tumor anoma

Fig. 3. Received scattered si

Layers Re

Coupling

TIES FOR HUMAN HEAD MODEL AND MOR

ric nt

Electric Conductivity Density

(S/m) (kg/m)

7 0.595 1030

8 0.244 1850

9 1.21 1043

l for elements of antenna array.

Signals ain signals are received and the antenna array. The received ed by four components as shown l from transmitter elements of

gnal from skull and brain layers, aly and the clutter [13].

gnal at antenna array element.

flections

Tumor Reflections

Clutters


572

III. DICTIONARY CREATIONElectromagnetic imaging (EMI) problem

as a dictionary selection problem. Thereforecreated by discretizing the target space. Cdiscretization results in a finite set of posspositions = {1, 2, .. , N}, where N iresolution and each position i is a 3D locatall processing examples, zi is fixed and is the

A. Temporal Compressed Sensing Preprocessing is performed on the receive

Fig. 3, for the subtraction of background andresulting in the signal shown in Fig. 4. It waare approximately only 25% of samples, winformation. Considering the signal to besamples for each element of the antenna arrasamples are selected for the analysis, givingcompressed sensing. This results in signa(=0.25) for each element. This preprocessall L elements of the antenna array to syobservation signal described by M samples as

M , , , LT Where T indicates transpose operation.

Fig. 4. Received signal after background and other re

B. Spatial Compressed Sensing The signal synthesizing described in Eq.

10% of the total N discrete spatial locationrandomly in the from the target space. Lineimplemented to evaluate the data for the rest resulting in full dictionary MN with N ofor full target space each described by M samcompressed sensing saves considerable ticollection. The resultant dictionary is then problem image reconstruction of the target sp

IV. COMPRESSED SENSING BASED INCompressed sensing (CS) has been widely

applications to recover signals and imag

N is formulated here

e, the dictionary is Consequently, the sible tumor target is the target space tion [xi, yi, zi]. For same for all N.

ed signal shown in d other reflections. as found that there which carry useful described by -ay, only the useful g rise to temporal

al with -samples ing is repeated for ynthesize a single s:

(1)

flections subtraction.

1 is done for only s that are selected

ear interpolation is of the target space,

observation signals mples. This spatial ime for the data

n tested in inverse pace.

NVERSION y used in numerous ges using limited

number of samples or measprinciples underlying CS are sptarget space to be reconstructedapplication valid.

The CS theory proves that fthe CS recovery algorithm canspace xN1, which maps relatively limited number of me

b

Where M1 represents the system is ill-posed and underdeof signal sparsity, xN1 careconstruct the target space, widesigned recovery algorithm. T(MIP) is a widely used framewThis requires the mutual incohesmall as possible. The mutual is defined as:

max

Where i and j denote the ith respectively.

Different CS recovery algthem, the greedy search algoritfor practical usage. As a canoOMP algorithm has receivedsimplicity and competitive recoit has been shown that the Oreconstructing both sparse andalgorithm is therefore implemenproblem solution, and the OMPto those for standard back-projethe column, which has the mcurrent residuals at each step is

V. SIMULATThe simulations are perfor

tumors of different shapes aalgorithm, as shown in Fig. 5atumor target is a combination oThe measurements vector bobservation signals, since it itargets at discrete spatial positisuperposition valid.

The reconstructed imagesmultiple tumors using stantechnique are shown in Fig. 6a compressed sensing (CS) recoalgorithm are shown in Fig. 7images are normalized to their a logarithmic 40-dB scale. The with CS techniques provides suin terms of resolution and accur

surements [6, 7]. The central parsity and incoherence [6]. The d here is sparse, which makes CS

for a given dictionary MN, n reconstruct the K-sparse target

the tumor distribution from a easurements vector bM1 by:

x (2)

amount of noise. Although the etermined, the prior information an be implemented to perfectly ith high probability, via properly

The Mutual Incoherence Property work for the CS recovery [14]. erence of the dictionary to be as incoherence of the dictionary

x, (3)

and jth column of the dictionary,

orithms are available. Among thm receives significant interest nical method in his family, the

d special attention due to its onstruction performance. In fact, OMP algorithm is reliable for

d near-sparse signals [15]. OMP nted in this paper for the inverse P resulting images are compared ection (SBP) technique. In OMP maximum correlation with the selected [16].

TION RESULTS rmed for a single and multiple

and sizes to test the proposed a and 5b, respectively. The test of the 1-millimeter point targets. b is a superposition of the is assumed here that the point ions do not interact, making the

corresponding to single and ndard back-projection (SBP) and 6b, respectively. Moreover,

onstruction results by the OMP 7a and 7b, respectively. All the

own maxima and are shown on results show that reconstruction

uperior images compared to SBP, racy.


573

(a)

Fig. 5. Target space with (a) sin

(a)

Fig. 6. SBP reconstructed i

(a)

Fig. 7. CS reconstructi

(b)

ngle and (b) multiple tumors of different shapes and sizes. Images us

(b)

images for (a) single tumor and (b) multiple tumors. Images use a 40

(b)

on for (a) single tumor and (b) multiple tumors. Images use a 40-dB

se a 40-dB scale.

0-dB scale.

B scale.


574

VI. CONCLUSIONS A novel method of nearfield EMI is proposed and tested for

the detection of tumors in the brain. A phantom of the human head is simulated in the forward problem, assuming the use of four-element antenna array to transmit electromagnetic signals and to record the received scattered time-domain signals. A dictionary is created, implementing both spatial and temporal compressed sensing (CS)

The images for the target space with single and multiple tumors of different shapes and sizes are reconstructed successfully. The inverse problem implements the OMP algorithm, and the results are much better than those reconstructed using SBP techniques. This shows the validity of the proposed algorithm. However, the reconstruction appears to provide the skeleton of the tumor shape, rather than the shape itself. Image processing can be used to enhance the quality of reconstructed images.

ACKNOWLEDGMENT This research work is partially supported by the National

Plan for Science and Technology at KACST, Kingdom of Saudi Arabia, under project number: 10-ELE996-02. The authors also acknowledge the support of the College of Engineering Research Center and the Deanship of Scientific Research at King Saud University in carrying out this research work.

REFERENCES [1] M. Benedetti, M. Donelli, G. Franceschini, M. Pastorino, and A.

Massa, "Effective Exploitation of the a Priori Information through a Microwave Imaging Procedure Based on the SMW for NDE/NDT Applications," IEEE Transactions on Geoscience and Remote Sensing, vol. 43, pp. 2584-2592, 2005.

[2] B. J. Mohammed, D. Ireland, and A. M. Abbosh, "Experimental Investigations into Detection of Breast Tumour Using Microwave System with Planar Array," IET Microwaves, Antennas & Propagation, vol. 6, pp. 1311-1317, 2012.

[3] M. Ostadrahimi, P. Mojabi, S. Noghanian, L. Shafai, S. Pistorius, and J. LoVetri, "A Novel Microwave Tomography System Based on the Scattering Probe Technique," IEEE Transactions on Instrumentation and Measurement, vol. 61, pp. 379-390, 2012.

[4] E. C. Fear, J. Bourqui, C. Curtis, D. Mew, B. Docktor, and C. Romano, "Microwave Breast Imaging with a Monostatic Radar-Based System: A Study of Application to Patients," IEEE

Transactions on Microwave Theory and Techniques, vol. 61, pp. 2119-2128, 2013.

[5] B. J. Mohammed, A. M. Abbosh, S. Mustafa, and D. Ireland, "Microwave System for Head Imaging," IEEE Transactions on Instrumentation and Measurement, vol. 63, pp. 117-123, 2014.

[6] E. Candes and J. Romberg, "Sparsity and Incoherence in Compressive Sampling," Inverse problems, vol. 23, p. 969, 2007.

[7] R. G. Baraniuk, E. Cands, R. Nowak, and M. Vetterli, "Compressive Sampling," IEEE Signal Processing Magazine, vol. 25, pp. 12-13, 2008.

[8] M. N. Tabassum, I. Elshafiey, and M. Alam, "Innovative Nearfield Electromagnetic Imaging System," in Proceedings of IEEE International Conference on Smart Instrumentation, Measurement and Applications (ICSIMA), Kuala Lumpur, Malaysia, 25-27 November, 2014.

[9] Computer Simulation Technology-CST. Documentation of CST Microwave Studio. Available: https://www.cst.com/Products/CSTMWS. Accessed in 2014.

[10] S. M. Yacoob and N. S. Hassan, "FDTD Analysis of a Noninvasive Hyperthermia System for Brain Tumors," Biomedical engineering online, vol. 11, p. 47, 2012.

[11] C. Gabriel, "Compilation of the Dielectric Properties of Body Tissues at RF and Microwave Frequencies," U.S. Air Force Report AFOSR-TR-96, http://transition.fcc.gov/oet/rfsafety/dielectric.html.

[12] I. Elshafiey, A. F. Sheta, M. Aldhaeebi, M. Alzabidi, and Z. Siddiqui, "Optimization of UWB Applicator for Hyperthermia Treatment of Human Head," in The 82nd ARFTG Microwave Measurement Conference 2013, pp. 1-5.

[13] Z. Haoyu, A. O. El-Rayis, N. Haridas, N. H. Noordin, A. T. Erdogan, and T. Arslan, "A Smart Antenna Array for Brain Cancer Detection," in Antennas and Propagation Conference (LAPC), 2011 Loughborough, 2011, pp. 1-4.

[14] D. L. Donoho and X. Huo, "Uncertainty Principles and Ideal Atomic Decomposition," IEEE Transactions on Information Theory, vol. 47, pp. 2845-2862, 2001.

[15] N. Aravind, K. Abhinandan, V. Acharya, and D. Sumam, "Comparison of OMP and SOMP in the Reconstruction of Compressively Sensed Hyperspectral Images," in International Conference on Communications and Signal Processing (ICCSP), 2011, 2011, pp. 188-192.

[16] T. T. Cai and W. Lie, "Orthogonal Matching Pursuit for Sparse Signal Recovery with Noise," IEEE Transactions on Information Theory, vol. 57, pp. 4680-4688, 2011.


575

/ColorImageDict > /JPEG2000ColorACSImageDict > /JPEG2000ColorImageDict > /AntiAliasGrayImages false /CropGrayImages true /GrayImageMinResolution 200 /GrayImageMinResolutionPolicy /OK /DownsampleGrayImages true /GrayImageDownsampleType /Bicubic /GrayImageResolution 300 /GrayImageDepth -1 /GrayImageMinDownsampleDepth 2 /GrayImageDownsampleThreshold 2.00333 /EncodeGrayImages true /GrayImageFilter /DCTEncode /AutoFilterGrayImages true /GrayImageAutoFilterStrategy /JPEG /GrayACSImageDict > /GrayImageDict > /JPEG2000GrayACSImageDict > /JPEG2000GrayImageDict > /AntiAliasMonoImages false /CropMonoImages true /MonoImageMinResolution 400 /MonoImageMinResolutionPolicy /OK /DownsampleMonoImages true /MonoImageDownsampleType /Bicubic /MonoImageResolution 600 /MonoImageDepth -1 /MonoImageDownsampleThreshold 1.00167 /EncodeMonoImages true /MonoImageFilter /CCITTFaxEncode /MonoImageDict > /AllowPSXObjects false /CheckCompliance [ /None ] /PDFX1aCheck false /PDFX3Check false /PDFXCompliantPDFOnly false /PDFXNoTrimBoxError true /PDFXTrimBoxToMediaBoxOffset [ 0.00000 0.00000 0.00000 0.00000 ] /PDFXSetBleedBoxToMediaBox true /PDFXBleedBoxToTrimBoxOffset [ 0.00000 0.00000 0.00000 0.00000 ] /PDFXOutputIntentProfile (None) /PDFXOutputConditionIdentifier () /PDFXOutputCondition () /PDFXRegistryName () /PDFXTrapped /False

/CreateJDFFile false /Description > /Namespace [ (Adobe) (Common) (1.0) ] /OtherNamespaces [ > /FormElements false /GenerateStructure false /IncludeBookmarks false /IncludeHyperlinks false /IncludeInteractive false /IncludeLayers false /IncludeProfiles true /MultimediaHandling /UseObjectSettings /Namespace [ (Adobe) (CreativeSuite) (2.0) ] /PDFXOutputIntentProfileSelector /NA /PreserveEditing false /UntaggedCMYKHandling /UseDocumentProfile /UntaggedRGBHandling /UseDocumentProfile /UseDocumentBleed false >> ]>> setdistillerparams> setpagedevice