Research On The Key Technologies Of Medical Image Registration In The Image-Guided Radiation Therapy

At present,cancer is a major killer of human health,and radiotherapy is one of the most effective ways of cancer treatment.Image-guided radiation therapy has become an important development trend of radiation therapy technology.In the process of image-guided radiation therapy implementation,medical images are acquired and used to register with medical image data which is used to make the treatment plan to monitor and correct the treatment errors of the radiation process quickly and accurately,so as to guide the whole process of radiotherapy and to ensure the accuracy and efficiency of radiotherapy,which has become an important impact on the effects of radiotherapy.Therefore,how to improve the accuracy and efficiency of the key technologies of medical image registration simultaneously and effectively in the image-guided radiation therapy,in order to effectively control the target radiation accuracy and shorten the treatment time of radiotherapy,it is very important to improve the cure rate of cancer.In this dissertation,the key technologies of medical image registration include medical image registration based on mutual information,2D-3D medical image registration and medical image non-rigid registration in the image-guided radiation therapy,which are researched from the point of view of accuracy and efficiency to correct the target migration error,position error and tissue deformation error and improve the accuracy of treatment results and the speed of treatment procedure in the image-guided radiation therapy.The research works of this dissertation are mainly summarized as follows:(1)For the problem that the object function is easy to get into local optimization due to much local extremes in the mutual information registration method,a multi-resolution medical image registration method based on the combination of a firefly algorithm and Powell algorithm is proposed in this dissertation.Firstly,a mutual information medical image registration algorithm based on firefly algorithm is put forward in this method.The normalized mutual information is used as the similarity measure and the registration parameters are expressed by the locations of fireflies in the algorithm,and mutual information function values are calculated according to the locations of fireflies and are set as brightness values of fireflies,and the best registration parameters are retrieved by updating the brightness and attractiveness iteratively while the mutual information function reaches the maximum value.The experimental results indicate that this algorithm can effectively overcome the problem that the mutual information function is easy to fall into local optimization,and the precision of registration result is improved obviously.Secondly,in order to further improve the accuracy and efficiency of the registration algorithm,we combine Powell algorithm and the firefly algorithm to construct a multi-resolution registration algorithm for medical images.The experimental results show that the proposed algorithmachieves sub-pixel level registration accuracy in comparison with existing mutual information registration algorithms.(2)For the problem of the slow speed and the large migration error of target contour in the clinical treatment requirements of image-guided radiation therapy,a parallel medical image registration algorithm of mutual information based on an adaptive inertia weight firefly algorithm and computing unified device architecture is proposed in this dissertation.The algorithm uses firefly optimization strategy based on dynamic adaptive inertia weight to find the optimal registration parameters.Since the adaptive inertia weight is introduced in the firefly algorithm,the firefly algorithm has strong global and local search ability.By adding random disturbance to improve the diversity of the population in each iteration,our algorithm effectively improves the problem that the mutual information function with more extreme is easy to fall into local optimum,thus improving the accuracy of registration results.After that,mutual information is calculated in parallel on the GPU using hierarchical model of CUDA and shared memory,and the spanning tree merging algorithm and instruction strategies are employed to further optimize the computation efficiency of mutual information.Experimental results show that our algorithm effectively utilizes the hardware architecture and the parallel computation ability of GPU and improves the execution speed of image registration based on mutual information algorithm.(3)For the problem of the position error and the long time in the fractional radiotherapy for clinical treatment requirements of the image guided radiation therapy,we propose a parallel 2D-3D medical image registration algorithm based on computing unified device architecture and the combinatorial similarity measure.Firstly,the generating process of DRR image in parallel is implemented using CUDA model in the algorithm;Then,the square difference of absolute value and pattern intensity are combined as a new similarity measure which is computing in parallel on GPU;Finally,the new combination similarity values are transfer to the CPU to find the optimal registration parameters using the fruit-fly optimization algorithm based on bacterial chemotaxis behavior.The experimental results show that the proposed algorithm can effectively improve the execution speed of the 2D-3D medical image registration process compared with the existing algorithms;Meanwhile,the proposed algorithm improves the accuracy of registration results adopting combinatorial similarity measure compared with the other similarity measures.(4)For the problem of the treatment error caused by tissue deformation and the long treatment time in the image-guided radiation therapy,a B-spline non-rigid medical image registration algorithm based on CUDA is proposed in this dissertation.First,the logarithm squared difference is considered as the similarity metric in the B-spline registration algorithm to improve registration precision.After that,we create a parallel computing strategy and look-up tables according to the local characteristics of B-splines transformation to reduce the complexity of the B-spline registration algorithm.As a result,the computing time of threetime-consuming steps including B-splines transformation,similarity metric computation and the analytic gradient computation of similarity metric,is efficiently reduced,for the B-spline registration algorithm employs the nonlinear conjugate gradient optimization method.Experimental results of registration quality and execution efficiency on the large amount of medical images show that our algorithm achieves a better registration accuracy in term of the differences between the best deformation fields and ground truth and an effective speedup ratio due to the parallel computing strategy on GPU.The research results of this dissertation can provide theoretical and technical support for further research and practice of some applications related to the image guide radiation therapy.Meanwhile,it has important theoretical significance and applicable value for improving the accuracy and efficiency of the image-guided heavy ion radiation therapy and accelerating the development of heavy ion radiation therapy technology.