Using Wavelet-Based Features to Identify Masses in Dense Breast Parenchyma

Abstract

Automated detection of masses on mammograms is challenged by the presence of dense breast parenchyma. The aim of this study is to investigate the feasibility of wavelet-based feature analysis in identifying spiculated and circumscribed masses in dense breast parenchyma. The method includes an edge detection step for breast border identification and employs Gaussian mixture modeling for dense parenchyma labeling. Subsequently, wavelet decomposition is performed and intensity as well as orientation features are extracted from approximation and detail subimages, respectively. Logistic regression analysis (LRA) is employed to differentiate spiculated and circumscribed masses from normal dense parenchyma. The proposed method is tested in 90 dense mammograms containing spiculated masses (30), circumscribed masses (30) and normal parenchyma (30). Free-response receiver operating characteristic (FROC) analysis is used to evaluate the performance of the method, achieving 83.3% sensitivity at 1.5 and 1.8 false positives per image for identifying spiculated and circumscribed masses, respectively.