Classification of Breast Mass Abnormalities using Denseness and Architectural Distortion

This paper presents an electronic second opinion system for the classification of mass abnormalities in mammograms into benign and malignant categories. This system is designed to help radiologists to reduce the number of benign breast cancer biopsies. Once a mass abnormality is detected and marked on a mammogram by a radiologist, two textural features, named denseness and architectural distortion, are extracted from the marked area. The denseness feature provides a measure of radiographic denseness of the marked area, whereas the architectural distortion feature provides a measure of its irregularity. These features are then fed into a neural network classifier. Receiver operating characteristic (ROC) analysis was conducted to evaluate the system performance. The area under the ROC curve reached 0.90 for the DDSM database consisting of 404 biopsy proven masses. A sensitivity analysis was also performed to examine the robustness of the introduced texture features to variations in sizes of abnormality markings.