| We have been developing an automated detection algorithm for masses on digital mammograms, but some larger areas remain as false-positive candidates. In this paper, we propose a simple method based on edge analysis by using a Prewitt filter to improve detection performance in terms of eliminating false-positive candidates. Edge patterns extracted within normal mammary-gland regions tend to be complex ones because of the presence of funicular components. On the other hand, patterns within malignant mass shadows tend to be white, plane surfaces owing to their high x-ray absorption ratio. By comparing these different features between masses and normal patterns, false-positive masses were eliminated. As a result of incorporating this method into our previous detection scheme, the number of false-positives decreased from 0.93 to 0.78 per image while maintaining the same true-positive rate. It was concluded that our improved scheme is effective for eliminating false mass candidates basically including edge patterns with larger mass area. |
