A fast and accurate method to register medical images using Wavelet Modulus Maxima

摘要

This paper presents a fast, accurate and automatic method to register images of rigid bodies. It uses wavelets to obtain control points. Wavelets are not shift invariant but the structures determined by the wavelet high pass image, the Modulus Maxima Image, provide the information necessary for a fast-rough convergence. These structures represent shapes from which we segment the invariant shapes for the images being registered. For example, the MRI and CT images of the brain can be considered as rigid bodies that do not undergo a change in shape over reasonable periods of time. By using a convex hull, we make the procedure insensitive to the internal changes in the object. Hence, even with the growth of tumors, the procedure registers brain images very accurately. The method uses the correlation coefficient to measure the similarity between images and to determine how well the images are registered. The method has been extensively tested with various types of images and in all cases the registration accuracy is very high. The correlation coefficient used to validate the registrations has deficiencies that occasionally required a visual inspection to terminate the algorithm after a point of marginal improvement.