Classification of breast lesions based on quantitative measures of tumor morphology

High resolution digitized mammograms were used in the development of a technique to classify breast lesions based on their boundary characteristics. Two different boundary descriptors were used to classify the tumors as benign or malignant: a measure of tumor circularity (μ_R/σ_R), and a measure of surface roughness. The surface roughness measurement was calculated as the percentage of angles with multiple boundary points. The greater the percentage of angles with multiple boundary points the more irregularly shaped the boundary. 94% of the tumors were automatically segmented to produce boundaries consistent with an outline drawn by an expert reviewer. A classification model which includes both descriptors was diagnostic with an estimated area under the ROC curve of 0.94 (se=0.04)