Automated Analysis of Mammographic Densities

Although interactive methods, such as the one just discussed, provide a continuous, quantitative scale for density assessment and have been shown to be quite reliable, they are still somewhat subjective and they are also labor intensive. For this reason it would be desirable to develop more objective techniques of characterizing mammographically dense tissue. To achieve this goal, several investigators have attempted to identify features that can be calculated automatically and on a continuous scale [19-24]. Mammographic density is distinguished in the image on the basis of both brightness and texture variations. Features to be used for image segmentation can be based on the image itself, on transforms of the image, or on the gray-level histogram.

Two features that have been successfully applied are a measure of image brightness variations based on the skewness (third moment) of the histogram of pixel gray levels and one of texture quantified by the fractal dimension. These measures have been found to be well correlated with subjective assessments of mammographic density [19]. Furthermore, these parameters have been demonstrated to be robust to simulated variations in mammographic technique, at the level of those found in clinical practice. This is important, because, in practical applications of risk prediction from mammo-graphic density measurement, images will be acquired over a period of several years and at different geographical sites.

0 0

Post a comment