Combining data from different algorithms to segment the skin-air interface in mammograms

This paper presents a method for combining several different estimates of the mammographic skin-air interface in order to eliminate noise inherent to each individual segmentation algorithm. Given that each algorithm provides a binary mask of the breast, the first step is to isolate pixels adjacent to the skin-air interface. A final estimate of the skin-air interface for each point results from the combination of skin-air interface location data from each procedure. Data for each point is grouped as a set, upon which statistical operators, such as the elimination of outliers, are applied. Since the skin-air interface is a continuous line, data from prior points is also used as an estimate of points that follow. Results are evaluated in terms of success with the combination of two skin-air interface segmentation algorithms. The resulting `hybrid´ technique overcomes several problems that beset each individual algorithm