Research On Deformation Registration Algorithm And Its Application In Adaptive Radiotherapy

At present,the treatment of malignant tumors relies heavily on radiotherapy technology.However,during radiotherapy,due to errors in positioning and organ deformation,its position and shape change during each treatment,even if in the small organ deformation,the target area may not receive enough dose and the organ at risk will be involved in the high dose area.The adaptive radiotherapy technology can effectively solve the above situation by combining the fractional image information to diagnose the patient.Therefore,precise radiotherapy poses new challenges to image registration technology,and exploring high-precision medical image deformation registration technology has important practical significance in clinical practice.In this paper,the key similarity measurement part in the current deformation registration technology is deeply studied,the characteristics of head and neck CT images are used to design experiments,the traditional registration algorithm is improved,and the dose accumulation analysis is carried out using the improved algorithm.The dose changes in the target area and each vital organ were clarified.The main research work of this paper is as follows:（1）According to the characteristics of head and neck CT images,normalized mutual information is selected as the similarity measure,and the multi-level B-spline model is used as the deformation model for registration,which reduces the grayscale caused by the small overlapping area of the images to a certain extent.The difference calculation error can achieve a good deformation registration effect between the whole for images with small deformation.（2）When registering images with large deformation or high local distortion,it is easy to fall into local extreme values,etc.,and it is impossible to simulate a smooth deformation field well,and even the phenomenon of deformation field folding occurs.Therefore,in this paper,the original target The Forbenius norm of the parameter matrix was introduced into the function as a regular term,and a normalized mutual information deformation registration method was studied,which resulted in a smoother deformation field and enhanced the robustness of image deformation registration.Awesome.The radiotherapy image data images of head and neck cancer patients were selected for registration experiments,3D modeling was performed for the target volume and organs at risk before and after registration,and the degree of overlap before and after registration under different algorithms was compared.The Dice similarity coefficients（DSC）of the PTV target area,left and right parotid glands obtained by experimental geometric evaluation in mutual information registration,normalized mutual information registration and the algorithm in this paper are: 0.88±0.04,0.68±0.07,respectively.,0.65±0.07,of which the Dice coefficient obtained under the algorithm in this paper reached the maximum value of 0.92;the Hausdorff distance value was also lower than other comparison algorithms,and the surface deviation range distances were all within the clinical error range.（3）Finally,the cumulative dose application of adaptive radiotherapy for head and neck was studied.The radiation dose data of patients with head and neck cancer were selected for experiments,and the differences between the cumulative dose parameters of each fractional dose volume histogram in radiotherapy and the cumulative dose parameters after image deformation registration were studied.After rigid registration,the mutual information registration algorithm and the registration algorithm proposed in this paper are used for dose accumulation.3D Slicer software is used to analyze different algorithms.In dose evaluation,HI value of uniformity index in rigid registration algorithm,mutual information registration algorithm and the algorithm proposed in this paper are: 1.157,1.149,1.123,the algorithm in this paper also reaches the minimum value of 1.123;In addition,gamma ray was used to quantify the dose-volume difference between two images.The difference of gamma index in the algorithm in this paper was 52.73%,which significantly reduced the dose-volume difference and provided a certain reference for the evaluation of registration quality in bed environment.