Hyun Hee Jo1, Helen Hong1
1Division of Multimedia Engineering,
To preserve tumor volume in DCE-MR breast images, we propose a demon-based deformable registration with rigidity constraint and density correction. First, the breast skin is extracted by using maximum gradient profile searching and the other breast tissues are classified into fat, muscle, glandular tissue and tumor using k-means clustering. Then the density of each breast tissue except tumor region is corrected by using histogram matching. Second, the tumor is localized in the subtracted images and is segmented in post-contrast enhanced images. Finally, tumor regions are rigidly transformed by averaging the magnitudes of deformation vector fields in narrow band and the other breast tissues are deformed by using demon-based deformable registration. As a result, the proposed deformable registration significantly reduces the effect of movement artifacts in subtracted contrast-enhanced images as well as efficiently preserves the tumor volume. Our deformable registration can be used for distinguishing benign lesions from malignancies and monitoring therapy.