Large deformation consistent elastic image registration

A new large deformation image registration algorithm, Large Deformation Consistent Elastic Image Registration (LDCEIR) algorithm, is proposed. The new algorithm is driven by the intensity difference of the template image T(x) and the target image S( x), and is based on the linear elastic model.; To accommodate large deformations, the algorithm solves the N-dimensional image registration in an (N + 1)-dimensional domain. A periodic sequence of N-dimensional images is constructed by inserting intermediate images between images T(x) and S( x), and a series of N-dimensional incremental transformations is estimated that deforms each intermediate image to the next image in the sequence. Large deformations from one image to the other are accommodated by concatenating the small-deformation incremental transformations. To achieve the inverse consistency, the LDCEIR algorithm jointly estimates a consistent set of incremental transformation functions.; The LDCEIR algorithm formulates the image registration problem as an optimization problem and uses the Fourier series expansion to parameterize the displacement fields of the transformations. A multi-resolution strategy is used in both spatial and frequency domains to iteratively estimate the Fourier series coefficients and the gradient descent method is used to search for the optimum solution.; The algorithm is characterized on phantoms, MRI brain images and CT inner ear images. The effect of varying the parameters used in the algorithm is analyzed and guidelines for adjusting the parameters under different situations are discussed. The algorithm is compared with the small deformation linear elastic registration algorithm and the large deformation viscous fluid registration algorithm theoretically as well as experimentally, where the small deformation elastic algorithm is contained in the new algorithm as a special case. The experiments show that the LDCEIR algorithm has much smaller inverse consistency error while the intensity error is still comparable to the fluid registration algorithm.