Abstract: A novel method is presented for efficient local asymmetric regions detection in medical images using image registration and mathematical morphology.First,we extend self-reflexive rigid binary image registration to find the symmetry plane of a gray-scale medical image.Then based on the registered image, a fast conditional erosion operator is used to adjust the border of bones and brain tissues to remove the likely symmetric region.These two steps are iteratively repeated using a mult-i resolution scheme to extract the final interesting area. The experiment to detect cholesteatoma in human head CT based on the proposed method shows encouraging results,and 95％ of the asymmetric regions are correctly detected.