Statistical based 3-D image registration approach with external markers

Abstract

In this paper, a new method for 3-D image registration based on the positions of corresponding external point markers in two modalities is presented. The proposed approach is based on the Bayesian theorem, the normal likelihood function (expressing the residuals) and the exponential priors to approximate the posterior density of the registration transformation. The transformation is expressed in terms of 9 parameters: 3 magnifications, 3 translations and 3 rotation angles. The parameters are estimated using the Markov Chain Monte Carlo sampling approach i.e. Gibbs sampling. The performance of the proposed registration technique is evaluated quantitatively using Monte-Carlo simulation techniques by generating images of the external markers in two modalities. Having transformed the corresponding marker positions to the same coordinate system using the estimated transformation, their residuals are computed. Thus, the various factors that affect the registration's accuracy can be investigated using the residuals of the external markers and the registration parameters. These factors include: the number of external markers and the positioning errors of the markers.