Design of Microwave-based Brain Tumor Detection Framework with the Development of Sparse and Low-Rank Compressive Sensing Image Reconstruction

Cancer is one of the leading causes of death, and the brain is one of the body’s cancer-prone organs.  The early detection of brain tumors can reduce cancer risk, which is practically assisted and conducted using scanners such as computed tomography (CT) and magnetic resonance imaging (MRI). However, those modalities are high-cost and large-sized, and they have a side effect risk to health. Alternatively, microwave imaging offers a novel cancer scanning method for early detection with low cost, small size and low health risk. Consequently, this research designs and creates a framework with a novel microwave image reconstruction algorithm inside. The framework is a component of the controller and image reconstructor for a portable microwave-based brain tumor detector that is open source and multi-platform. For the novel algorithm, this research proposes a CS-based imaging algorithm by exploiting the data‘s sparse and low-rank properties. The experiment shows that the proposed algorithm can give better qualitative and quantitative reconstruction results compared to a full-sampling-based as well as CS-based algorithm.

Compressive sensing; Framework; Image reconstruction; Low-rank; Microwave imaging; Sparse

In Cancer is the second-leading cause of death globally. According to the WHO, the death rate due to cancer reached 9.6 million in 2018 (WHO, 2018). The brain is one of the organs susceptible to cancer. The early detection of brain tumors is essential to mitigate the risk of cancer. There are many examples of tumor detection using imaging technologies such as X-ray, computed tomography (CT), magnetic resonance imaging (MRI) and positron emission tomography (PET) (Gao and Jiang, 2013). However, these modalities still have disadvantages such as radiation level, device complexity, operational costs and size (Chandra et al., 2015). Therefore, researchers develop a new modality by utilizing microwaves (Dilman et al., 2017; Shtoda et al., 2017).

        Microwave imaging has used various algorithms to reconstruct the scan results, such as time-domain inversion (Ali and Moghaddam, 2010), ultra-wideband magnitude combined tomography (UPC), algebraic reconstruction technique (ART) (Elevani et al., 2016, Elevani et al.,2017[HR1] ), simultaneous algebraic reconstruction technique (Aprilliyani et al., 2017), filtered back projection (FBP) (Ramadhan et al., 2017; Ramdani et al., 2018) and the algorithm based on compressive sensing (CS) (Basari and Ramdani, 2019).  However, UPC, ART and FBP algorithms have disadvantages because they require much data to get decent results. Meanwhile, the compressive sensing approach reduces the amount of data needed for reconstruction because it uses sampled data. The data also tend to be sparse in a frequency domain, that is having many elements containing zero value among the total data set (Donoho et al., 2016; Basari and Ramdani, 2019). Previous research proposed CS with total variation (TV) regularization solved by the alternating direct method of multipliers (ADMM) (Razzak et al., 2019). The method shows a smoother reconstructed image and lower mean square error (MSE) than SART and sparse CS.

The research contribution lies in the proposed image reconstruction algorithm for the software aspect and the proposed framework for the portable hardware aspect. The proposed algorithm improves image reconstruction results by using the low-rank property of the data. It combines sparse and low-rank properties of the data based on the compressive sensing approach. The idea is that the microwave measurement data are not only sparse but also low rank. The low-rank property is represented by imposing a nuclear norm (the sum of a matrix's singular value). It has been utilized for facial recognition (Luo et al., 2014), feature extraction (Yang et al., 2017), hyperspectral unmixing (Giampouras et al., 2016), CT image reconstruction (Yang et al., 2017) and MRI reconstruction (Ulas et al., 2016) due to its robustness. The data are said to be low rank because data matrices tend to be highly correlated. In an implementation, the nuclear norm is imposed on the CS optimization problem to consider low ranking.

Researchers have developed simple, low-cost and portable medical devices (Hugeng and Kurniawan, 2016).  This research introduces the design of a framework that is universal, open-source and multi-platform concerning the hardware aspect. The framework is for a controller and for packaging the developed image reconstruction algorithm. It can run on a card-sized computer, such as the Raspberry Pi, to be used as a portable brain tumor detector component. The framework can be operated as if a user operates a portable brain tumor detector.

The rest of this paper is organised as follows. Section 2 describes the details of the proposed method in this paper. Section 3 shows parameter settings, results and our respective analyses. Finally, Section 4 concludes this study.

This research has successfully designed and implemented a framework for controller and image reconstructor components of a universal, open-source, multi-platform, portable microwave-based brain tumor detector. The framework has been implemented in the Python language using Python libraries that support scientific computing. It can run on Raspberry Pi, a card-sized computer platform. Frameworks can be operated in the simulation mode as appropriate for the user and can function according to predetermined specifications.

Algorithm development shows that low rankness by nuclear norm can be applied as regularization in microwave image reconstruction under the compressive sensing (CS) approach. Compared to the CS method added with TV, which has given the best results compared to the SART and FBP methods, the proposed method can give the same results – and perhaps even better ones.

Qualitatively, the proposed SLR?CS algorithm shows the same image reconstruction in color and size to differentiate tumor and tissue. Quantitatively, this method can provide a better similarity and error value to the reference image, measured by SSIM and MSE parameters. The SLR?CS method provides a 45% similarity rate (SSIM) and an 18% pixel error rate (MSE), which is a 1% increase from previous studies (CS added with TV).

        The authors acknowledge Universitas Indonesia’s support through the Q3 Research Grant 2020 and Q1Q2 Research Grant 2019 under contract number NKB?0309/UN2.R3.1/HKP.05.00/2019.

Ali, M.A., Moghaddam M., 2010. 3D Nonlinear Super-Resolution Microwave Inversion Technique using Time-Domain Data. IEEE Transactions on Antennas and Propagation, Volume 58(7), pp. 2327–2336

Aprilliyani, R., Prabowo, R.G., Basari., 2017. Comparison Analysis between SART and ART Algorithm for Microwave Imaging. PhotonIcs & Electromagnetics Research Symposium - Fall (PIERS - FALL),

Basari., Ramdani, S., 2019. Evaluation on Compressive Sensing-based Image Reconstruction Method for Microwave Imaging. PhotonIcs & Electromagnetics Research Symposium - Spring (PIERS-Spring)

Chandra, R., Zhou, H., Balasingham, I., Narayanan, R.M., 2015. On the Opportunities and Challenges in Microwave Medical Sensing and Imaging. IEEE Transactions on Biomedical Engineering, Volume 62(7), pp. 1667–1682

Copper Mountain Technologies, 2020. Vector Network Analyzer Automation, Available Online at: https://coppermountaintech.com/automation/, Accessed on May 20, 2020

Dilman, I., Akinci, M.N., Cayoren, M., Akduman, I., 2017. Differential Microwave Imaging of the Stroke-affected Brain via Diffraction Tomograph. The 25^th Telecommunications Forum TELFOR

Donoho, D.L., 2006. Compressed Sensing. IEEE Transactions on Information, Volume 52(4), pp. 1289–1306

Elevani, D., Ramadhan, R., Parastika, D.T., Basari., 2016. Evaluation of Algebraic Reconstruction Technique Algorithm for Microwave Imaging. Electromagnetic Research Symposium (PIERS), Shanghai

Elevani, D., Ramadhan, R., Parastika, D.T., Basari., Sanpanich, A., 2017. On the Performance of Algebraic Reconstruction Technique Algorithm for Microwave Imaging. IEEE 5^th Asia?Pacific Conference on Antennas and Propagation (APCAP), Taiwan

Gao, H., Jiang, X., 2013. Process on the Diagnosis and Evaluation of Brain Tumors. Cancer Imaging, Volume 13(4), pp. 466–481

Guo L., Bialkowski K., Abbosh A., 2017. Compressive Sensing using Approximated Total Variant Transform to Reduce the Required Number of Antennas in Radar-Based Medical Imaging. IEEE International Symposium on Antennas and Propagation & USNC/URSI National Radio Science Meeting, USA

Giampouras, P.V., Themelis, K.E., Rontogiannis, A.A., Koutroumbas, K.D., 2016. Simultaneously Sparse and Low-Rank Abundance Matrix Estimation for Hyperspectral Image Unmixing. IEEE Transaction on Geoscience and Remote Sensing

Hugeng, H., Kurniawan, R., 2016. Development of the ‘Healthcor’ System as a Cardiac Disorders Symptoms Detector using an Expert System based on Arduino Uno. International Journal of Technology, Volume 7(1), pp. 78?87

Luo, L., Yang, J., Qian, J., Yang, J., 2014. Nuclear Norm Regularized Sparse Coding. 22^nd International Conference on Pattern Recognition, pp. 1834?1839

Muntasa, A., 2017. Windowing System Facial Detection based on Gabor Kernel Filter, Fast Fourier Transform, and Probabilistic Learning Vector Quantization. International Journal of Technology, Volume 8(1), pp. 196?208

Nusantara, A.C., Purwanti, E., Soelistiono, S., 2016. Classification of Digital Mammogram based on Nearest-Neighbor Method for Breast Cancer Detection. International Journal of Technology. Volume, 7(1), pp. 71?77

Qaisar, S., Bilal, R.M., Iqbal, W., Naureen, M., Lee, S., 2013. Compressive Sensing: From Theory to Applications, a Survey. Journal of Communications and Network, Volume 15(5), pp. 443?456

Ramadhan, R., Prabowo, R.G., Aprilliyani, R., Basari., 2017. Comparison Analysis between Filtered Back Projection and Algebraic Reconstruction Technique on Microwave Imaging. International Symposium on Biomedical Engineering (ISBE), Bali, Indonesia

Ramdani, S., Astyani, A.P., Basari., 2018. Filtered Back Projection and Simultaneous Algebraic Reconstruction Technique for Image Formation on Square-shaped Physical Phantom Aimed at Microwave Imaging Applications. PhotonIcs & Electromagnetics Research Symposium (PIERS-Toyama)

Ranson, R.G., 2015. Measurement Automation using Python. ARMMS RF & Microwave Society

Raspberry Pi Foundation, 2020.  Raspberry Pi 3 Model B+. Available Online at:  https://www.raspberrypi.org/products/raspberry-pi-3-model-b-plus/, Accessed on April 3, 2020

Razzak, I.H., Rizkinia, M., Basari., 2019. Image Reconstruction based on Compressive Sensing using Total Variation Spatial Regularization for Microwave Imaging. Photonics & Electromagnetics Research Symposium, Rome, Italy

Shtoda, D.O., Arkusha, Y.V., Mustetsov, M.P., 2017. The Microwave Imaging Method for Express Diagnostic of Cancer. IEEE First Ukraine Conference on Electrical and Computer Engineering (UKRCON), Ukraine

Ulas, C., 2016. Spatio-temporal MRI Reconstruction by Enforcing Local and Global Regularity via Dynamic Total Variation and Nuclear Norm Minimization. IEEE 13^th International Symposium on Biomedical Imaging (ISBI), Prague, Czech Republic

World Health Organization (WHO), 2018. Cancer. Available Online at: https://www.who.int/news-room/fact-sheets/detail/cancer, Accessed on April 3, 2020

Yang, K., 2019. Nonlocal Weighted Nuclear Norm Minimization Based Sparse?Sampling CT Image Reconstruction. IEEE 16^th International Symposium on Biomedical Imaging (ISBI)

Yang, W., 2017. A Nuclear Norm-based Matrix Regression-based Projections Method for Feature Extraction. IEEE Access, Volume 6