Mammographic Registration Method Based on Optical Flow and MultiresolutionComputing

摘要

Mammography is a potent weapon in the fight against Breast Cancer, due in largepart to its widespread availability and low cost. Despite the fact that mammography can detect small lesions as early as two years before they become palpable on physical exam, between 10 and 30 percent of cancerous lesions go undetected during evaluation by the radiologist. One approach to improving detection rates involves comparing mammograms of the same breast from successive years. Since most forms of breast cancer develop slowly, multiple view techniques might be able to detect subtle changes indicative of cancerous growth. This thesis proposes a computer aided system designed to bring two images into correspondence, or alignment, so that they can be compared and evaluated for possible abnormalities. The system estimates a mapping between two images by calculating the optical flow, or apparent intensity change, between a source and target mammogram. The efficiency of the proposed registration system is enhanced by utilizing a multiresolution approach whereby images are compared at more than one scale. Preliminary results suggest the potential usefulness of this system as part of a clinical computer aided detection (CADx) system.