A Wavelet-Based Similarity Measure to register pre- /intra-operative MR images of the brain

Abstract

Definition of a proper similarity measure to be adapted to a specific application is a crucial step in registration of medical images. The problem with most commonly used similarity measures in medical applications is that they perform registration in spatial domain based on simplifying assumptions about the interdependencies of the pixel intensity values. Therefore, they are incapable of decorrelating spatially-varying intensity inhomogeneity, occurring in MR imaging. To overcome this problem, Residual Complexity (RC) has been introduced for correcting intensity inhomogeneity in the Discrete Cosine Transform(DCT) domain. In this work, it is proposed to employ Discrete Wavelet Transformation(DWT) instead of DCT which is more efficient in representing the final residual image between the two registered images with minimum compression complexity. Here, the performance of Wavelet based RC (WRC) is compared to RC and some well-known similarity measure. WRC shows more than 30% improvement in the registration result in comparison with RC .