Optimized 2D/3D Medical Image Registration using the Estimation of Multivariate Normal Algorithm (EMNA)

X. Yang (Switzerland), W. Birkfellner (Austria), and P. Niederer (Switzerland)


2D3D registration, optimization, estimation of evolution algorithm, medical image processing In this paper, we propose that the transformation variables are mutually dependent and prove that the t


Recently, a new type of iterative optimization algorithm Estimation of Distribution Algorithms (EDAs) was put forward [10]. The idea behind this approach is to use estimation and sampling to replace the crossover and mutation of Genetic Algorithms (GA). In each generation, the population (i.e. a collection of points) is replaced by a probability distribution that explicitly models that population. The covariance matrix of the search distribution reflects correlated parameters (non separability), and its application is especially suitable for ill conditioned and non-separable problems and for problems where the cost function exhibits some long narrow valleys. Dependencies of transformation parameters utilized in 2D/3D registration are firstly described, and a probabilistic distribution model of the transformation parameters is derived. Second, the Estimation of Multivariate Normal Algorithm (EMNA) is introduced for 2D/3D registration. A number of tests were performed using simulated data. The results demonstrate that EMNA is a potential powerful candidate for an optimized 2D/3D registration process.

Important Links:

Go Back