Adaptive Grid Generation Based Non-rigid Image Registration using Mutual Information for Breast MRI

Abstract

In this paper a new approach for non-rigid image registration using mutual information is introduced. A fast parametric method for non-rigid registration is developed by adjusting divergence and curl of an intermediate vector field from which the deformation field is computed using finite-central difference method. Mutual information is newly employed as the similarity measure in the gradientbased cost minimization (or mutual information maximization) of the existing registration framework. The huge amount of data associated with MRI is handled by a fully automated multi-resolution scheme. The adaptive grid system naturally distributes more grids to deprived areas. The positive monitor function disallows grid folding and provides a mean to control the ratio of the areas between the original and transformed domain. The flexibility of the adaptive grid allocation could dramatically reduce processing time with quality preserved. Mutual information facilitates robust registration between different image modalities. Different types of joint histogram estimation are compared and integrated with the system. This scheme is applied on dynamic contrast-enhanced breast MRI, which requires the registration algorithm to be non-rigid, contrastenhanced features preserving. Preliminary experiments show promising results and great potential for future extension.

Evaluation of a Nonrigid Motion Compensation Technique Based on Spatiotemporal Features for Small Lesion Detection in Breast MRI

“…Although there has been a significant amount of research on nonrigid motion compensation techniques in brain imaging, few methods have been so far proposed for breast MRI. Most proposed techniques employ physically motivated deformation models [6,7], transformations based on the deformation of Bsplines, [8,9], elastic transformations [10], and more recently adaptive grid generation algorithms [11]. We present in this paper a novel elastic image registration method based on the variational optical flow computation and will study its impact on the shape of the enhancement curves for small lesions.…”

Velocity-Based Moving Mesh Methods for Nonlinear Partial Differential Equations

“…Further, [32] combines the algorithm with a least-squares finite element approach to solving the divcurl system (3.18) directly for ν(x,t), and apply the resulting algorithm to problems where the boundary deforms. The algorithm has also been applied in three space dimensions [91], to track moving shock waves [87], and in image registration [41]. Delzanno et al [51] however note some situations where the quality of the grids obtained compares unfavourably with a Monge-Kantorovich transformation approach.…”

Medical image registration based on feature and mutual information

“…When the both images achieve the best registration, the grayscale MI of corresponding pixels approaches the maximum [7,8].…”

“…In order to reduce the changing sensitivity of MI to overlapping parts, Studholme [9] and Maes [10] separately proposed a presentation style which is normalized MI. They are shown in the formula (7) and (8). (9) Where ij P represents the joint probability of feature point sets A P and B P , that is, at the same time the probability of extracting i P from A P and extracting j P from B P .…”