Comparison between parzen window interpolation and generalised partial volume estimation for nonrigid image registration using mutual information

摘要: Because of its robustness and accuracy for a variety applications, either monomodal or multimodal, mutual information (MI) is very popular similarity measure (medical) image registration. Calculation MI based on the joint histogram two images to be registered, expressing statistical relationship between intensities at corresponding positions. However, calculation not straightforward. The discrete nature digital images, sampled as well in intensity spatial domain, impedes exact histogram. Moreover, during registration often an will sought non grid position floating image. This article compares common estimators context nonrigid multiresolution medical registration: Parzen window interpolator (IIP) generalised partial volume estimation (GPV). Starting from BrainWeb data realistic deformation fields obtained patient experiments show that GPV more robust, while IIP accurate. Using combined approach, average error 0.12 mm intramodal 0.30 intermodal achieved.