Mutual Information Driven Inverse Consistent Nonlinear Registration

A fully symmetric nonlinear viscoelastic image registration method, under the demons paradigm is developed. The symmetric cost function includes mutual information (MI) as a similarity measure, regularization of the transformation, and inverse consistent constraint (ICC). Alternative strategy is used to minimize the divergent terms with different properties in the cost function to avoid the difficulties in balancing between the performance of registration and low Inverse Consistency Error (ICE). The diffeomorphism of the registration is achieved by the composition scheme. Validation studies show that the proposed method provides sub-voxel ICE. The improvement of registration performance is also demonstrated.