Registration of Multiple Medical Images Using N-dimensional Mutual Information

Abstract

The medical image registration has been all confined in registering two images and rarely involved multiple images (three and more than three). Using the expanded mutual information measure (E-MIM) to register multiple images, which is inefficiency, can't meet the clinical requirement, and mutual information (MI) values are not necessarily nonnegative. In this paper, we introduce an N-dimensional mutual information measure (N-MIM), which can ensure MI values are nonnegative, bounded to range from 1 to 2. At the same time, the rate of the registration has moved up. Then this definition is tested and proved to be effective on registration of multiple lumbar vertebra images through simulation.