The robust, hierarchical registration of multimodal medical images

There is a large clinical advantage to using imaging data from more than one modality at a time (e.g., MR, IR, CAT, PET, Ultrasound), informing reliable patient assessments. There is a corresponding urgent clinical need to reliably register images arising from different modalities. Conventional registration programs are unable to register these images with any reliability. However, by maximizing the Mutual Information (MI) between the images, it is possible to register images from different modalities. Improved performance results by registering at successive scales of spatial resolution. However, current methods are computationally demanding. The present work introduces the following innovations: (i) Unlike all versions in the literature, the MI criterion is formulated for continuous image domains; (ii) MI maximization is expressed as a variational principle. A regression method for the solution of the linearized variational equation is proposed.