Registration of Brain MRI/PET Images Based on Adaptive Combination of Intensity and Gradient Field Mutual Information

摘要

Traditional mutual information (MI) function aligns twomultimodality images with intensity information, lacking spatialinformation, so that it usually presents many local maxima that canlead to inaccurate registration. Our paper proposes an algorithm ofadaptive combination of intensity and gradient field mutualinformation (ACMI). Gradient code maps (GCM) are constructed bycoding gradient field information of corresponding original images.The gradient field MI, calculated from GCMs, can providecomplementary properties to intensity MI. ACMI combines intensity MIand gradient field MI with a nonlinear weight function, which canautomatically adjust the proportion between two types MI incombination to improve registration. Experimental resultsdemonstrate that ACMI outperforms the traditional MI and it is muchless sensitive to reduced resolution or overlap of images.