Information Intensity-based Research On Optimization Of Medical Image Registration

In recent years,with the development of computer science and medical imaging technology,Medical Image Processing has become a hot research field.As a basic and key technology of Medical Image Processing,Medical Image Registration,which is a key step to realize medical image analysis,complete medical image understanding,and assist clinical diagnosis and treatment,has had a very wide range of applications in the medical field,and the research on it has theoretical and practical significance.This thesis first introduces and summaries relevant researches at home and abroad,then elaborates on the definition,principle,classification,registration methods,and other basic theories of Medical Image Registration.It elaborates on the four main steps of image registration: image interpolation,space transformation,similarity measurement,and optimization strategy.This thesis concludes that the image registration method based on information intensity is the current mainstream method.The principle of this type of method is to measure the similarity or difference between two images and define the measurement function.The core is to find an algorithm that both its convergence and accuracy have very good optimization.Secondly,this thesis introduces an improved differential evolution algorithm--Bernstein Search Differential Evolution Algorithm(BSA).This algorithm has a simple structure and good running speed.However,this algorithm has the disadvantages of difficulty to find the peak value in the later iteration of the medical image registration,and the initial iteration convergence speed is too slow.Based on this,this thesis draws on the idea of backpropagation and proposes an improved differential evolution algorithm--Poisson search differential evolution algorithm(PSA)and normal search differential evolution algorithm(NSA).Compared with the BSA algorithm,these two improved differential search algorithms,especially NSA,can be better adapted to medical image registration and take into account the breadth and breadth of the search.It is easy to find the global optimal solution,and the optimization process is more stable and efficient.Finally,this paper designs four modules of MATLAB experiment includes initialization,image preloading,algorithm startup,and experimental result storage.This experiment uses single-mode lung COVIDx-CT and multi-mode brain CT,MR image Retrospective Image Registration Evaluation Project(RIRE)as the data set,CCRE as the information intensity similarity measure,and BSA,PSA,NSA,and PSO as the optimization strategies,among which PSO is the industry benchmark algorithm.The experimental results show that the improved differential search algorithm NSA and PSA proposed in this paper show higher registration accuracy and faster convergence speed than BSA and PSO when performing single-mode/multi-modal medical image registration,and it also improves the performance of the medical image registration optimization algorithm.The main research tasks of this thesis are:(1)Use BSA as an optimal strategy for medical image registration;summarize the advantages and disadvantages of BSA based on the performance of it in the process of medical image registration;(2)Based on the characteristics of information intensity-based medical image registration,this thesis draws on the idea of backpropagation,modified the shortcomings of BSA,and proposed PSA and NSA.(3)This thesis conducts single-modality and multi-modality medical image registration experiments between PSA,NSA and compares them with industry benchmark particle swarm optimization algorithms(PSO)and BSA.The result is: PSA and NSA can not only overcome local optimization problems,but can improve the accuracy of registration,increase the utilization rate of the population,and speed up the convergence speed of finding the extreme value.