Research On Multi-modal Medical Image Registration Algorithm Based On Mutual Information Correlation Ratio

Medical image registration,which plays a very important role in image fusion,clinical diagnosis,surgical navigation,surgical planning and postoperative evaluation,and so on,is a very important technology in medical image processing.With the rapid development of medical imaging technology,multi-modal medical image registration plays a more significant role in many clinical applications,so it crucial to investigate multi-modal image registration.To obtain accuracy and robustness multi-modal image algorithm,image preprocessing,image similarity measures,optimization,regularization technology have been studied in this paper.The research work and innovative achievements of this paper are mainly outlined as follows:(1)In order to remove speckle noise in optically related tomographic images,three denoising methods,anisotropic diffusion filters,wavelet transforms,and bilateral filters,are introduced in detail.Experimental results show that the three denoising methods can all achieve good denoising effect and the bilateral filters achieve the best result,which retains image edge information as well as removes noise(2)Mutual information is the most important similarity measure in multi-modal image,however,it only measures the dispersion of the joint density based on the assumption of a correspondence of intensity classes between two images and ignores the functional mapping of intensity values.To resolve this question,a mixed similarity measure called correlation ratio-based mutual information(CRMI)has been used combing B-spline free form deformation and improved stochastic gradient optimization.Comparing CRMI with other two algorithms on the clinical data,the new measure need less registration time and have higher registration accuracy.,(3)Spatial regularization is essential in image registration,which can help to avoid both physically implausible displacement fields and local minima during optimization.Conventional non-rigid registration methods assume a continuous and smooth deformation field throughout the image.However,discontinuities exist when thoracic and abdominal organs such as liver or lungs with sliding motion during respiration.In this paper,total variation(TV)is used as the penalty term to preserve the discontinuous boundaries.The average target registration error on public data is reduced to I.27mm and superior(10%to 40%)to other techniques.The proposed method has been demonstrated with a more credible displacement field near the discontinuous interface and more accuracy registration results in the clinical data.