Medical Image Registration by Maximizing a Hybrid Normalized Mutual Information

Normalized mutual information (NMI) has developed into a quite effective measure for rigid and multimodality medical image registration. However, the robustness of the measure is questionable. One possible reason for this is the interpolation artifacts which are introduced in the determination of the joint histogram. In the present paper, a novel exponential scheme is proposed to construct the joint histogram. The new scheme can yield a more smooth NMI function. Furthermore, considering the robustness and the speed, we advise to register images using a complementary manner called Hybrid-NMI. Experimental results show that the advised Hybrid-NMI method can improve the registration speed and robustness without significantly sacrificing registration accuracy.