Detection method for architectural distortion based on analysis of structure of mammary gland on mammograms

Architectural distortions, as well as clustered microcalcifications and masses, are important findings in interpreting breast cancer on mammogram. We have developed an automated detection algorithm for distorted areas based on the concentration of mammary glands in order to aid physicians with diagnosis. The purpose of this study is to suggest an improvement to our previous method to achieve higher sensitivity. The extraction method of the mammary gland was changed to employ both a shape index and curvedness so as not to be influenced by the background density from the breast thickness. The directionality of a normal of mammary gland is toward the nipple whereas that in an abnormal gland is toward another, distorted area. Normal mammary glands were eliminated by virtue of their conformation with the normal directionality. Our image database consisted of 117 cases with architectural distortions. The first-stage detection rate and the number of correct detections of ROIs were improved with our method. It was concluded that our detection method would be effective. (c) 2005 Published by Elsevier B.V.