Symmetry of Projection in the Quantitative Analysis of Mammographic Images

An important practical question in the analysis of mammo-graphic density is the extent to which information about mammographic pattern is carried by any one of the four views of a typical mammographic exam. This applies to both symmetry between the right (R) and left (L) breast in a given projection, and to symmetry of projection, i.e., between cranial-caudal (CC) and medial-lateral oblique (MLO) projections of the same breast. If the information available from a single image is representative, then the measurement of quantitative, objective parameters could be simplified and the additional work and cost of digitization, storage, and analysis of the other three views in a mammographic exam could be eliminated. Invariance of breast parenchyma characterizations between views in a mammographic exam has implications in retrospective studies of mammographic parenchyma. For example, when studies include women diagnosed with breast cancer in one breast, it is sometimes necessary to view images from the other breast to avoid observer bias.

There is good evidence to suggest that a high degree of symmetry normally exists. A study of more than 8000 women, by Kopans et al. [38], found that diagnostically significant asymmetries were reported in only 3% of mammographic exams. In a study by Boyd et al. [39], exact agreement in Wolfe grade classification was observed between RCC and LCC projections in 71% of the 78 pairs of films classified.

To evaluate symmetry for measures of mammographic density, we carried out a study on 90 sets of patient mammograms spanning the full range of mammographic density [40]. For each case, comparisons were made between projections (RCC vs. RMLO), and between images of the left and right breast (RCC vs. LCC) for (1) subjective classification of mammographic density by radiologists (SCC); (2) interactive density thresholding (projected breast area A and percent mammographic density PD); (3) regional skewness measurement; and (4) fractal texture measurement.

For subjective classification (SCC), very high levels of concordance (Spearman correlation Rs = 0.95) were obtained for comparisons of mammographic density between the two breasts. Similarly, all objective parameters tested showed strong left-right symmetry (RCC vs LCC). The slopes in each of the regressions include 1.0 and the intercepts include 0.0, within 95% confidence intervals. Although natural variation between the right and left breasts is expected, these differences do not significantly affect the value of either subjectively or objectively derived features from the mammograms. Strong correlations were also observed for comparisons of the RMLO and RCC projection. In most cases, the 95% confidence intervals for the slope and intercept include 1.0 and 0.0, respectively. This suggests that each of the parameters reflects the general organization of breast tissue, independent of projection.

These results indicate that a representative characterization of mammographic density can be obtained from analysis of a single projection of one of the breasts. Typically, the choice of which projection to use is arbitrary. In studies for which disease is present in one of the breasts, an image from the contralateral breast should be used.

0 0

Post a comment