Incorporating connected region labelling into automated image registration using mutual information

Abstract

The information theoretic measure of mutual information has been successfully applied to multi-modality medical image registration for several applications. There remain however; modality combinations for which mutual information derived from the occurrence of image intensities alone does not provide a distinct optimum at true registration. The authors propose an extension of the technique through the use of an additional information channel supplying region labelling information. These labels which can specify simple regional connectivity or express higher level anatomical knowledge, can be derived from the images being registered. The authors show how the mutual information measure can be extended to include an additional channel of region labelling, and demonstrate the effectiveness of this technique for the registration of MR and PET images of the pelvis.