Multimodality Medical Image Registration Based On Mutual Information

The geometric alignment or registration of images is a fundamental task innumerous applications in medical image processing. Medical diagnosis often benefitsfrom the complementarity of the information in images of different modalities. A greatbulk of registration algorithms occurred over the past decade. Among them imageregistration based on mutual information (MI) has become an increasing popular matchmetric due to its wide applicability and overall accuracy. In this paper, multimodalityimage registration based on mutual information is discussed mainly.The problem of local maxima in multimodality image registration based on mutualinformation is discussed in this paper. Artifact caused by image noise and interpolationis analyzed firstly. Filter preprocessing based on hamming window is introduced todecrease the local maxima. Simulations have been done to illustrate that local maximaare eliminated to a great extent by using low-pass filter preprocessing. optimizationalgorithm is introduced to search the global maxima for the registration parameters. Amodified PSO (Particle Swarm Optimization) algorithm is adopted to search the properparameters. The experiment results demonstrate that this method performs fairly well.Finally, local maxima in non-rigid image registration based on mutual informationis discussed in this paper. Image gradients and mutual information are used to be theregistration metric. The local rigid registration is performed to produce correspondingpoints automatically. Thin plate interpolation is used to realize the global elasticregistration. A pyramid approach for multimodality image registration is presented.Simulations have been done to illustrate that the efficiency and accuracy of this methodin registration strategy.