Premium
Skin artifact removal technique for breast cancer radar detection
Author(s) -
Caorsi S.,
Lenzi C.
Publication year - 2016
Publication title -
radio science
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.371
H-Index - 84
eISSN - 1944-799X
pISSN - 0048-6604
DOI - 10.1002/2016rs006011
Subject(s) - radar , artifact (error) , finite difference time domain method , computer science , signal (programming language) , breast cancer , artificial neural network , artificial intelligence , cancer , telecommunications , medicine , optics , physics , programming language
In this paper we propose a new model‐based skin artifact cleaning technique with the aim to remove skin reflections with good effectiveness, without introducing significant signal distortions, and without assuming a priori information on the real structure of the breast. The reference cleaning model, constituted by a two‐layer geometry skin‐adipose tissue, is oriented to all the ultrawideband radar methods able to detect the tumor starting by the knowledge of each trace recorded around the breast. All the radar signal measurements were simulated by using realistic breast models derived from the University of Wisconsin computational electromagnetic laboratory database and the finite difference time domain (FDTD)‐based open source software GprMax. First, we have searched for the best configuration for the reference cleaning model with the aim to minimize the distortions introduced on the radar signal. Second, the performance of the proposed cleaning technique has been assessed by using a breast cancer radar detection technique based on the use of artificial neural network (ANN). In order to minimize the signal distortions, we found that it was necessary to use the real skin thickness and the static Debye parameters of both skin and adipose tissue. In such a case the ANN‐based radar approach was able to detect the tumor with an accuracy of 87%. By extending the performance assessment also to the case when only average standard values are used to characterize the reference cleaning model, the detection accuracy was of 84%.