Local image registration a comparison for bilateral registration mammography

摘要

Early tumor detection is a key factor in reducing breast cancer mortality. Screening mammography is the most widely available method for early cancer detection. However, the large amount of images to be read at each radiological session stress the radiologist ability to process each case with uniform attention. Computer aided detection (CADe) systems can improve tumor detection rate; but the current efficiency of these systems is not adequate and the correct interpretation of CADe outputs requires more time from the experts. Computer aided diagnosis systems (CADx) are being designed to improve cancer diagnosis accuracy and reduce radiological workload; but the efficiency CADx in screening mammography is not yet adequate for clinical use. Our hypothesis is that CADx sensitivity for mammography applications can be enhanced by considering the natural symmetry between the right and left breast. The primary objective of this work is to evaluate co-registration algorithms for the accurate alignment of the left to right breast for CADx applications. Three different set of registrations algorithms were tested: One based on B-Spline deformable transformations and two based on Demon's free form deformations. A set of mammograms was artificially altered to create a ground truth set of 132 images for the evaluation of the registration efficiency. The registration accuracy was visually inspected and quantitatively evaluated using mean square error, mutual information and correlation metrics. The results indicated that the B-Spline deformable registration outperforms Demon's for the task of registering mammography images.