Finding CLS Using Multiresolution Oriented Local Energy Feature Detection

Abstract

In this paper, we present a novel technique for the detection of the curvilinear structures (CLS) in a mammogram based on a multiresolution, oriented local energy analysis. Local energy enables the detection not only of linear structures; but also features of several different kinds in a unified framework. It is possible to distinguish between such feature types using the local phase. In a separate post-processing stage, the behaviour of energy over multiple scales can be used to determine a) whether a response is due to a feature or to noise and b) to estimate at each location the local width of a CLS. Orientation information computed from steerable filters is used in the same post-processing stage to distinguish between curvilinear structures and speck-like responses such as microcalcifications which, on a micro-scale, resemble CLS. By combining scale, phase and orientation information we can distinguish the CLS from non-CLS locally linear features as well as localised structures with high gradients and thus remove only the CLS whilst leaving the remaining important image information intact.