Registration-Based Interpolation Using a High-Resolution Image for Guidance

A method is presented for interpolation between neighbouring slices in a grey-scale tomographic data set, when a high-resolution image of the same patient is also available. Spatial correspondence between adjacent slices in the high-resolution image are established using a voxel-based non-rigid registration algorithm. These spatial correspondences are used to calculate a set of vectors, which are transferred from the high-resolution image to the lower resolution image by rigidly registering the image volumes together. Linear interpolation is then carried out along these vector directions between neighbouring slices in the low-resolution image. This method has been compared to standard linear interpolation, shape-based interpolation and registration-based interpolation in MR head volumes. Results using a mean square difference error measure show that the proposed method outperforms the other three interpolation techniques.