Multimodal medical image registration using particle swarm optimization with influence of the data's initial orientation

Imaging techniques are an excellent example for the continuous improvement of medical possibilities by technical innovation. One particularly relevant field of research is multimodal registration, where data sets must be aligned in order to make their structures overlay. The overlay of the image positions is reached by optimizing a similarity metric. A commonly used measure applied in this process is the normalized mutual information, which also will be used in this paper. Due to iteratively improving this function value, a transformation can be determined, which adapts the data in order to make the images overlap each other as accurately as possible. For this improvement process, different mathematical optimization methods are in use. One approach is Particle Swarm Optimization (PSO), a nature-inspired optimization technique. In the present work, four variants of the PSO algorithm are presented and applied to medical image data. Although classical PSO with standard parameters is shown to have some limitations, considerable improvement can be obtained by the modification of the calculation rules, the choice of the parameter values and the choice of the objective function. Our experimental results illustrate the substantial potential of PSO for this type of application.