Point-based medical image registration and error prediction considering noise perturbation

A novel point-based registration method for medical use considering noise perturbation was presented. The method is based on maximum likelihood esitimation (MLE), which is achieved by iteratively minimizing the absolute orientation problem (AOP) defined in the Mahalanobis space. The method can estimate not only the rigid transformation but also its variance based on the error propagation theory. The variance of the rigid transform is further propagated into the target registration error (TRE) and the expectation of the TRE is calculated for registration evaluation and fiducial marker (FM) configuration optimization. The numerical simulation shows the proposed method outperforms the classic AOP solution in both correctness and precision with the relative predictive error of the TRE less than 2%, of the standard deviation of the estimated transformation less than 4%.