Precise detection of early breast tumor using a novel EEMD-based feature extraction approach by UWB microwave

• Published in: Medical & biological engineering & computing Vol. 59; no. 3; pp. 721 - 731
• Main Authors: Liu, Guancong, Xiao, Xia, Song, Hang, Kikkawa, Takamaro
• Format: Journal Article
• Language: English
• Published: Berlin/Heidelberg Springer Berlin Heidelberg 01.03.2021; Springer Nature B.V
• Subjects: Algorithms; Backscattering; Biomedical and Life Sciences; Biomedical Engineering and Bioengineering; Biomedicine; Breast cancer; Computer Applications; Datasets; Empirical analysis; Feature extraction; Human Physiology; Imaging; Kernel functions; Original Article; Principal components analysis; Radiology; Support vector machines; Tumors; Ultrawideband; UWB; Feature selection; Early breast cancer detection; SVM; EEMD
• ISSN: 0140-0118; 1741-0444
• DOI: 10.1007/s11517-021-02339-5

The accurate detection of early breast cancer is of great significance to each patient. In recent years, breast cancer non-invasive detection technology based on Ultra-Wideband (UWB) microwave has been proposed and developed extensively, which is complementary to the existing methods. In this paper, a novel approach is proposed for tumor existence detection based on feature extraction algorithm. Firstly, the breast features are obtained by Ensemble Empirical Mode Decomposition (EEMD) and valid correlation Intrinsic Mode Function (IMF) selection. Secondly, raw feature datasets are constructed and then simplified by Principal Component Analysis (PCA) or Recursive Feature Elimination (RFE). Finally, the detection is realized by Support Vector Machines (SVM). The influence of different kernel functions and feature selection methods on detection results is compared. In this study, 11,232 sets of backscatter signals from simulation results of four different categories’ breast models are utilized. And feature dataset is constructed by 24 specific features from each signal’s four valid components. The results demonstrate that the proposed method can extract representative features and detect the early breast cancer effectively with the accuracy of 84.8%. Graphical abstract