Image Standardization for Digital Mammography

Abstract

Computer assisted diagnosis of mammography has received much attention is the last several years. A much cited problem of many of the developments is the use of a limited number of cases in the design and test databases. A second problem has surfaced recently. While CAD algorithms have relied on a single instrument for digitization of screen-film mammograms, CAD performance has been shown to depend on the source of the digital data. There are two approaches for this problem: (1) increase the robustness of the CAD algorithms; or (2) transform images so that they are valid for a particular CAD program. In [1], we presented an algorithm to standardize images, based on registering pairs of images using a genetic algorithm, allowing the measurement of the parameters for the geometric and intensity transformations. This potentially allows a more generalized application of CAD for different digitizers or different sensors as required for multicenter clinical trials and clinical application.