Automated analysis of mammographic densities.

Information derived from mammographic parenchymal patterns provides one of the strongest indicators of the risk of developing breast cancer. To address several limitations of subjective classification of mammographic parenchyma into coarse density categories, we have been investigating more quantitative, objective methods of analysing the film-screen mammogram. These include measures of the skewness of the image brightness histogram, and of image texture characterized by the fractal dimension. Both measures were found to be strongly correlated with radiologists' subjective classifications of mammographic parenchyma (Spearman correlation coefficients, Rs = -0.88 and -0.76 for skewness and fractal dimension measurements, respectively). Further, neither measure was strongly dependent on simulated changes in mammographic technique. Correlation with subjective classification of mammographic density was better when both the skewness and fractal measures were used in combination than when either was used alone. This suggests that each feature provides some independent information.