Research On Medical Image Registration Algorithm Based On Free Form Deformation Model And Multi-scale Feature

With the rapid development of medical imaging technology,various medical images have emerged.However,due to various differences between different sources of images or environmental changes caused by different time of image acquisition,it is difficult for researchers to integrate images accurately and completely for processing.To obtain comprehensive and complete image information,it is necessary to use image registration technology to match two or more images obtained under different acquisition equipment,different acquisition time or different acquisition Angle.At present,the image registration technology at home and abroad has been relatively mature in the field of rigid registration,but due to the nonlinear changes of human organs caused by muscle contraction,non-rigid medical image registration technology is more critical in clinical medicine.Non-rigid medical registration has become a hot topic in digital image processing at home and abroad.Based on the current domestic and international status and technical achievements of medical image registration,this paper studies the basic framework and mathematical description of image registration,and deeply studies its key technologies,including spatial transformation,interpolation algorithm and similarity.Measure,optimize algorithms,etc.,and classify and summarize the commonly used algorithms of each technology.The following work has been done on non-rigid medical image registration problems:1.For the medical images of human tissues prone to local deformation,a free deformation model registration algorithm based on B spline was proposed,and experimental analysis was carried out.It was found that although it could fit well for local deformation,it could not achieve better registration for large affine transformation globally,and registration failure was easy to occur.In order to solve this problem,this paper further put forward the combination of SIFT feature matching and VDSR registration algorithm of super-resolution reconstruction,first using the SIFT method of reference image and the floating image feature point extraction,the first step in the image by affine transformation of coarse matching,then coarse image multi-scale transformation after the match,the low scale image as B spline fitting method for local registration method of input,finally after super-resolution reconstruction to get the registration image of the original scale.Experiments show that this method not only improves the registration accuracy and robustness,but also improves the registration efficiency.2.Build the testing and evaluation platform of medical image registration algorithm.The platform not only includes image registration experiments of a variety of algorithms and evaluation methods for registration results,and displays the results in the form of intuitive graphics,but also includes simple image preprocessing operations.Before performing the registration algorithm,the registration image can be manually scaled,rotated and projected.The establishment of medical image registration platform improves the efficiency of experiment execution,reduces the complexity of experiment and improves the repeatability of experiment results.