IEEE Proof Web Version IEEE ANTENNAS AND WIRELESS PROPAGATION LETTERS, VOL. 13, 2014 1 Hybrid Artifact Removal for Confocal Microwave Breast Imaging M. A. Elahi, A. Shahzad, M. Glavin, E. Jones, and M. O’Halloran Abstract—Several factors determine the effectiveness of an early-stage artifact removal algorithm for the detection of breast cancer using confocal microwave imaging. These factors include the ability to select the correct time window containing the artifact, the ability to remove the artifact while being robust to normal variances, and ability to effectively preserve the tumor response in the resultant signal. Very few (if any) of the existing artifact removal algorithms incorporate all of these qualities. In this letter, a novel hybrid artifact removal algorithm for microwave breast imaging applications is presented, which combines the best attributes of two existing algorithms to effectively remove the early-stage artifact while preserving the tumor response. This algorithm is compared to existing algorithms using a range of appropriate performance metrics. Index Terms—Artifact removal, breast cancer, microwave imaging, ultrawideband radar. I. INTRODUCTION O NE OF the most important components of any confocal microwave imaging (CMI) system for breast cancer de- tection is the early-stage artifact removal algorithm. The early- stage artifact is composed of the input signal, the reflection from the skin–fat interface, and any antenna reverberation present. This artifact is typically several orders of magnitude greater than the reflections from any tumors present within the breast. There- fore, if the artifact is not removed effectively, it could easily mask tumors present within the breast. In this letter, a novel hybrid artifact removal algorithm for microwave breast imaging applications is presented. The nov- elty of this study is threefold. First, the authors propose a hybrid artifact removal algorithm that combines the best attributes of the Entropy-based Time Windowing algorithm [1] and Wiener Filter algorithm [2] to effectively remove the early-stage ar- tifact while preserving the tumor response. Second, the algo- rithm is evaluated using an anatomically and dielectrically ac- curate 3-D finite-difference time-domain (FDTD) model (com- pared to the 2-D homogeneous FDTD model originally used by Zhi and Chin [1]). Third, the hybrid algorithm presented here shows a very clear improvement compared to the original algo- rithm across a range of appropriate metrics. Manuscript received October 30, 2013; revised November 28, 2013; accepted January 02, 2014. Date of publication January 09, 2014; date of current version nulldate. This work was supported by Science Foundation Ireland under Grant No. 1l/SIRG/I2120. The authors are with the Department of Electrical and Electronic Engi- neering, National University of Ireland Galway, Galway, Ireland (e-mail: m.elahi1@nuigalway.ie). Color versions of one or more of the figures in this letter are available online at http://ieeexplore.ieee.org. Digital Object Identifier 10.1109/LAWP.2014.2298975 The remainder of the letter is organized as follows. Section II describes the proposed artifact removal algorithm in detail. Section III describes the 3-D numerical breast phantom used to evaluate the algorithm. Section IV details the various tests applied to the artifact removal algorithm and the corresponding results. Finally, the conclusions and suggestions for possible future work are discussed in Section V. II. ARTIFACT-DOMINANT WINDOW SELECTION FOR ADAPTIVE ARTIFACT FILTERING The proposed novel artifact removal algorithm combines the best attributes of the Entropy-based Time Window algorithm and the Wiener Filter algorithm. The first step of the proposed artifact removal algorithm is to automatically select the artifact-dominated portion of backscat- tered signals based on entropy values. In order to better se- lect the artifact window, the proposed algorithm improves upon the original Entropy-based Time Window artifact removal al- gorithm proposed by Zhi and Chin [1]. While their algorithm is effective in removing the artifact, it often fails to correctly estimate the exact portion of the signal containing the artifact. It also tends to remove part of tumor response when early-time artifact and tumor responses overlap in time. Therefore, an im- proved algorithm was developed by the authors that more accu- rately estimates the artifact-dominated portion of the signal. The Entropy-based Time Window algorithm is based on the assumption that the artifacts in the received signals are similar across all channels, unlike the case for the tumor response in real breast imaging scenarios (the tumor response is delayed and attenuated differently due to variations in the tissue structures at each channel). A larger value of entropy is obtained from similar artifacts in the early portion of the radar signal, and conversely, the tumor reflections result in a much lower entropy value. A window function estimated to contain the artifact can be defined based on the entropy values. First, a probability density function is created by normalizing each received radar signal (1) where is the received signal at the th channel and is the total number of channels. This equation satisfies and and can be interpreted as energy density in the antenna domain. The -order Renyi entropy at time sample is defined as (2) 1536-1225 © 2014 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.