These results show that increases in the level of breast tissue density as assessed by mammography are associated with increases in risk for breast cancer.

Quantitative classification of mammographic parenchyma based on radiological assessment has been shown to provide one of the strongest estimates of the risk of developing breast cancer. Existing classification schemes, however, are limited by coarse category scales. In addition, subjectivity can lead to sizeable interobserver and intraobserver variations. Here, we propose an interactive thresholding technique applied to digitized film-screen mammograms, which assesses the proportion of the mammographic image representing radiographically dense tissue. Observers viewed images on a CRT display and selected grey-level thresholds from which the breast and regions of dense tissue in the breast were identified. The proportion of radiographic density was then calculated from the image histogram. The technique was evaluated for the mammograms of 30 women and is well correlated (R > 0.91, Spearman coefficient) with a six-category subjective classification of radiographic density by radiologists. The technique was found to be very reliable with an intraclass correlation coefficient between observers typically R > 0.9. This technique may have a role in routine mammographic analysis for the purpose of assessing risk categories and as a tool in studies of the etiology of breast cancer, in particular for monitoring changes in breast parenchyma during potential preventive interventions.

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.