Nonrigid Medical Image Registration by Finite-Element Deformable Sheet-Curve Models
Image-based change quantitation has been recognized as a promisingtool for accurate assessment of tumor's early response tochemoprevention in cancer research. For example, various changeson breast density and vascularity in glandular tissue are theindicators of early response to treatment. Accurate extraction ofglandular tissue from pre- and postcontrast magnetic resonance(MR) images requires a nonrigid registration of sequential MRimages embedded with local deformations. This paper reports anewly developed registration method that aligns MR breast imagesusing finite-element deformable sheet-curve models. Specifically,deformable curves are constructed to match the boundariesdynamically, while a deformable sheet of thin-plate splines isdesigned to model complex local deformations. The experimentalresults on both digital phantoms and real MR breast images usingthe new method have been compared to point-based thin-plate-spline(TPS) approach, and have demonstrated a significant and robustimprovement in both boundary alignment and local deformationrecovery.