E K

FIGURE 3 Gray-level histogram of a digitized mammogram indicating thresholds iEDGE for segmentation of the area of the breast and iDY for the area of dense tissue. Adapted with permission from Byng JW, Boyd NF, Fishell E, et al. The quantitative analysis of mammographic densities. Physics in Medicine and Biology 1994; 39:1629-1638.

Pixels in regions of mammographic density have a gray level i>iDY. The area under the histogram above this threshold is representative of the projected area of mammographic density

FIGURE 3 Gray-level histogram of a digitized mammogram indicating thresholds iEDGE for segmentation of the area of the breast and iDY for the area of dense tissue. Adapted with permission from Byng JW, Boyd NF, Fishell E, et al. The quantitative analysis of mammographic densities. Physics in Medicine and Biology 1994; 39:1629-1638.

The use of gray-level thresholding to distinguish mammo-graphic density allows a relatively simple decision criterion to be applied, which has contributed to reproducible measures, both within and between observers, of both A and the extent of mammographic density within the breast.

To facilitate clinical studies in which PD is computed on a large number of mammograms, a Windows-based program has been developed. The program provides a user interface (Fig. 4) where intensity windowing can be applied to aid in visualizing the edge of the breast and the boundaries of regions of parenchymal tissue. Thresholds for the gray level corresponding to the ¿EDGE and ¿DY can then be set. Pixels corresponding to the thresholds appear in color — in this example red for ¿EDGE and green for ¿DY. In addition, there are tools available allowing the operator to use the mouse to describe regions of the image that are to be excluded from the histogram calculation. A multi-segmental line can be defined to segment the shadow of the pectoralis muscle adjacent to the chest wall. Similarly, polygonal areas can be outlined to eliminate the effect of radiographic markers, which may inadvertently superimpose on the image of the breast. After the thresholds have been set and the exclusion regions defined, the algorithm proceeds in a raster fashion, from the back of the breast to where the skin edge threshold is encountered, counting pixels lying within the image of the breast as well as those whose intensities lie above ¿DY. PD can then readily be calculated.

0 0

Post a comment