Detection of architectural distortion from the ridges in a digitized mammogram.

Architectural distortion (AD) has been described as a focal retraction of the breast tissue. Blood vessels, milk ducts and spicules in the breast tissue appear as ridges in the mammogram. We hypothesize that radiating ridges are an indicator of an AD site. Using a window-based approach, features derived from the ridges have been utilized in a radial basis function support vector machine to classify regions as containing or not containing AD. The classification is performed on the Mammographic Image Analysis Society (MIAS) database and on the Digital Database For Screening Mammography (DDSM). The proposed approach reports peak performance of a sensitivity of 90% (93%) at 26 (17) false positives per mammogram in the MIAS (DDSM) database. [ABSTRACT FROM AUTHOR]