Computer-aided detection of microcalcifications in mammography

摘要

A method of aiding detection of breast cancer by computer analysis of a mammogram image includes computing filtered second spatial derivative of intensity values at the pixels of the image, in the form of a Laplacian. The filtering is iterative adaptive smoothing of the first spatial derivative, which smoothing is applied to achieve relatively great smoothing effect where there is relatively little local variation in derivative value and a relatively small or no smoothing effect where there is relatively great local variation in derivative value. This has the effect of preserving locations of zero crossings in the Laplacian which correspond to edges or boundaries in the image. Regions of negative Laplacian value are labelled and connected. These labelled regions are locally bright spots in the image. From the preserved boundary locations of these locally bright spots, a plurality of feature measures are computed, indicative of their respective brightnesses, shapes and edge contrasts. Those locally bright spots having a weighted combination of logarithms of these feature measures exceeding a predetermined value are identified as microcalcifications of a type associated with cancer, and it is detected where such identified microcalcifications are grouped in clusters.