Medical Image Registration Research Based On Mutual Information And Demons Algorithm

With the development of the medical imaging technology, medical imaging technology of image registration plays an increasingly important role and become a research focus and hot spots in the recent years. Image registration is mainly used with the CT image, MRI image, PET(Positron Emission Tomography) for image matching, image fusion and other aspects of stereotactic radiotherapy.Medical image registration is finding a spatial transformation to achieve the corresponding points of the two images to reach consensus on space, so that making the two images are fusing and matching. However, in practical application the accuracy, speed, automation and robustness of medical image registration are still facing many difficulties and challenges. In our study, find that for the brain medical images elastic registration, the traditional method has drawback of having bigger calculation capacity, slower processing speed. To solve the problem, we have approached mutual information and Demons registration algorithm, and put forward a new method for brain medical images elastic registration. The experimental results show the effectiveness, stability and robustness of proposed algorithm. The main work and research results are as follows:(1) By considering the local contour and global changes of the image, we propose a medical image registration method based on active contour models, which combines cubic B-spline interpolation, active contour models, mutual information and LBFGS optimization algorithm for brain image to achieve image registration. The experimental results show that the new algorithm has better results compared with the traditional algorithm.(2) Demons and related series of classic registration algorithms are studied, we point out their weak points, such as computational complexity and lack direction information. Therefore, we propose a frequency-domain Demons registration processing method based on wavelet transform theory. The method uses wavelet transform that can realize the benefits of better location on each scale, orientation and location information. Image feature can be revealed by the high frequency, low frequency of image transformation. Experimental results show that the algorithm is effective and robust.