Mammogram Screening by Automated Followup: A Feasibility Study.

摘要

This report addresses the problem of image registration in eases where pixel-to-pixel correspondence can't be established. Specifically, the interest is in images containing 3-D elastic and nonstructured objects whose appearance varies with acquisition parameters. The work is motivated by the problem of mammogram screening based on comparison between mammograms of the same patient acquired in different screenings. Misregistration between temporally spaced screenings arises from minor differences in 3-D positioning and compression, as well as, normal changes in tissue that are function of time. The objective is to identify corresponding regions in two images, similarly to what is done by medical experts. The locations of the regions are determined based on the locations of identifiable landmark points, and each corresponding region's extent is determined by characteristics of the older mammogram. An image pair is covered with overlapping circular regions without gaps and the proposed algorithm provides for further splitting of larger regions. In order to gain an insight into the problem of mammogram misregistration, the work has been extended into problem of mammogram simulation. The developed simulation algorithms encompass the problems of modeling breast tissue, compression and X- ray image acquisition.