Medical Image Registration Based On Mutual Information And Wavelet Transformation

With the development of medicine and computer technology, medical imaging technology isplaying an increasingly important role in clinical application. The purpose of medical imageregistration is to do image geometric transformation for images of different periods and differentmodes and to make two or more images to exactly the same anatomical position. The results shouldmake the registration of two images match at all the anatomical points. The image registration basedon mutual information can be automated to achieve high accuracy by eliminating the need forimage preprocessing and it is the medical image registration trends.Firstly, in this paper the principle of medical image registration is introduced. Secondly, theregistration methods of image registration based on mutual information is researched. Finally themethods'advantages and disadvantages are analyzed. Based on the study of registration methods,this paper proposes some improvements:To solve the drawback of registration based on typical mutual information neglecting thespatial information of images, a new medical image registration based on combining mutualinformation with edge gradient similarity is developed. It was extracted the edge of image bycalculating the wavelet transform modular maximum. And the gradient similarity coefficients ofedge image was calculated and used to multiply by the mutual information to form the finalregistration metric. The method can reduce the number of local optimal solutions and get the globaloptimal solution by combining mutual information with edge gradient information of the images.Experimental results showed that the algorithm had high precision and effectiveness.To overcome the local maxima and the computing expansive shortcoming of registration basedon mutual information algorithms, wavelet based multi-resolution approach is applied inregistration. The sub-images decomposed by wavelet transformation are registered using PSO andthen the results are applied as the start point for the high-resolution image registration usingPowell's method. Experimental results show that the algorithm not only improves the speed ofimage registration and image registration accuracy is also reached the sub-pixel level.