Research on Voxel Indensity Based Multi-modal Medical Image Registration When Gross Outliers Are Observed
Zhuang Tian-ge · Hangtian yixue yu yixue gongcheng · 2004
Objective To find the most accurate and robust similarity measures for clinical use by comparison research on popular similarity measures for voxel intensity based multi-modal medical image registration, when gross outliers are observed in multi-modal medical images. Method Under theoretic analysis of different similarity measures used in registration of multi-modal medical images with gross outlier, using real multi-modal medical images without and with gross outliers and random noise signals, registration experiments were implemented to evaluate different performances of similarity measures for accuracy and robustness. Result It was found that normalized mutual information based registration method has special capability in getting accurate registration results when gross outliers are observed in multi-modal medical image registration, whereas correlation ratio and mutual information failed to get correct registration results. Conclusion Compared with other similarity measures, normalized mutual information was the most accurate and robust similarity measure for 3D multi-modal medical image registration when gross outliers are observed in images to be registered.