A Research On The Method Of Medical Image Registration

The medical image registration plays an important role in detection of disease, operation navigation and disease diagnosis, and it has emerged as one of key technologies in medical image. In this paper, The2D single-mode color medical image registration and3D multimodality medical image registration are mainly studied. In the process of single-mode color medical image registration, the color medical image is expressed as quaternion matrix, so the method of image registration is studied in the model of quaternion; In the process of3D multimodality medical image registration, ICA algorithm and Geometric algebra theory are applied to image registration,so the realisation of image registration is studied in the model of Geometric algebra.The emphases in medical image registration are summarized as follows:Firstly, According to the fact that color medical image registration is usually realized in labeling method, but in this paper the automatic method of single-mode color medical image registration is studied. The quaternion theory is applied to color image registration, and a measure of single-mode color medical image registration is proposed based on quaternion theory.The rotation amount is calculated by the principal axes of quaternion color medical image, and the translation amount is determined by the cross-correlation measure. For verify the method,The single-mode PET brain image and single-mode PET abdomen image are simulated and realized, and the registration result is good.The proposed algorithm is based on quaternion, the form of quaternion takes color image as a whole, makes more relevance between R, G, B, and has a relatively small amount of calculation.Secondly, the fast method of3D multimodality medical image registration is studied, for reducing the calculation amount, the image contour is extracted as registration set. The ICA algorithm is introduced to the registration method, According to independent statistics of signals, the independent component axis of two medical images is separated, The rotation amount is constructed by combining the independent component axis with Geometric algebra; the translation amount is calculated by the centroid of the two image contours. The Clinical medical image data coming from Vanderbilt university and the simulated brain data coming from the BrainWeb are matched by the proposed method, good results and high precision are achieved. The advantage of the methd is visual and vivid, the amount of calculation is less in the Geometric algebra.