Wavelets and Fuzzy Contours in 3D-CAD for Digital Breast Tomosynthesis

Abstract

Summary In this chapter we present a novel approach for the detection of different kinds of lesions in Digital Breast Tomosynthesis datasets. It consists in working directly on the projected views, providing the advantage of a reduced data volume to process, while staying independent of any given reconstruction algorithm, not yet fully optimized for this emerging modality. Our aim was to develop a single processing framework for the detection of different kinds of breast lesions. Introducing fuzzy processing enables us to maintain the evidence, and the strength of the evidence, gathered from each DBT projection image for each potential finding without making hard decisions in isolation. First, the projected views are filtered using different banks of multiscale wavelet filters allowing to better fit the wavelet to the pattern that may vary in a defined range of sizes. The different filter responses are then thresholded and the result is combined to obtain microcalcification and mass candidates. For each candidate, we create a fuzzy contour through a multi-level thresholding process. We extract attributes for each candidate contour that are characteristic for the different kinds of lesions to be detected. Fuzzy set definitions for the classes of the respective lesions are introduced that allow for the computation of fuzzy membership values for each candidate contour. Then, an aggregation operator is presented that combines information over the complete set of projected views, resulting in 3D fuzzy particles. A final decision is made taking into account information acquired over different successive processing steps. Visual examples of detection results are presented, along with a preliminary quantitative evaluation of the algorithm.