Finding curvilinear structures in mammograms

Abstract

We develop a model-based algorithm for the extraction of curvilinear structures in x-ray mammograms, corresponding to milk ducts, blood vessels, fibrous tissue, and spiculations. The diagnostic basis for the problem is sketched. Results of applying the algorithm are shown. The results accord closely with the CLS structures observed by Radiologists, and the algorithm has supported further processing to identify potential lesions and to match left and right breast x-ray pairs. A new perspective is provided on the debate on the relevant merits of film density images versus transmitted light.