Multi-Modal Medical Volume Registration Using a New Information Theoretic Measure

Abstract

This work presented a new 3D registration approach for multi-modal medical volumes. Our approach employed a generalized entropy called Arimoto entropy, which is a generalization of Shannon entropy. Our method proposed in this article applies the Jensen Arimoto divergence as a registration criterion and measures the similarity between the 3D medical volumes acquired from various modalities using this new criterion. The goal of this work is to obtain the maximum value of the new registration criteria by exploiting the quasi-Newton optimization scheme. Simultaneously, two volumes are completely registered, along with the final spatial transformation obtained. In order to evaluate our presented algorithm, the experiments on real three-dimensional medical volumes are designed and performed. Results of registration experiments illustrated that our approach is more effective and proves better registration accuracy. Additionally, a comparison with two classic measures based on information theory, cross-cumulative residual entropy and normalized mutual information, is carried out.