Image registration plays a pivotal role in many clinical applications. The misaligned or the "floating" image has to be accurately registered with the reference image. Similarity measures are used to register images by finding an accurate match between the reference image and the floating image. Accurate registration helps in better diagnosis of disease. Many similarity metrics have been proposed both for mono-modality and multimodality image registration. Among these metrics, mutual information based algorithms perform better for multimodality image registration applications. Mutual information is an intensity based metric and does not require any specification of landmarks. These properties make it a good similarity metric for multimodality image registration. Registration is achieved by maximizing the mutual information.

In this project we developed a scheme which combines gradient information, k-Means clustering and mutual information to improve the success rate of registration process. We present different ways to combine the information available and compare it to the existing methods.

Figure 1: Block showing different ways of combining information.

Figure 2: Reference and floating images. (a) Before Registration. (b) After Registration. The red circles show the misregistration in X.

This work was a collaborative effort with Prof. Lars Ewell in the Dept. of Radiation Oncology, Health Science Center, University of Arizona.

Publications:

1. Narendhran Vijayakumar, Lars Ewell, and Jeffrey J. Rodriguez, Baldassarre Stea, "Inferior Brain Lesions Monitored Using Diffusion Weighted Magnetic Resonance Imaging" (abstract), presented at the Sino-American Network for Therapeutic Radiology and Oncology (SANTRO) Symp., Aug. 28-30, 2008, Beijing, China.