A Multi-resolution CLS Detection Algorithm for Mammographic Image Analysis

Abstract

Curvilinear structures (CLS) are locally one-dimensional, relatively thin objects which complicate analysis of a mammogram. They comprise a number of anatomical features, most especially connective tissue, blood vessels, and milk ducts. The segmentation, identification and removal of such structures potentially facilitate a wide range of mammographic image processing applications, such as mass detection and temporal registration. In this paper, we present a novel CLS detection algorithm which is based on the monogenic signal afforced by a CLS physical model. The strength of the proposed model-based CLS detector is that it is able to identify even low contrast CLS. In addition, a noise suppression approach, based on local energy thresholding, is proposed to further improve the quality of segmentation. A local energy (LE)-based junction detection method which utilises the orientation information provided by the monogenic signal is also presented. Experiments demonstrate that the proposed CLS detection framework is capable of producing well-localized, highly noise-tolerated responses as well as robust performances as compared to classical orientation-sampling approach.