Image Registration Algorithm In Radiotherapy

The rapid development of modern medical imaging technology has promoted the development of radiotherapy.With the continuous progress of medical image processing and analysis,the precise location and tracking of tumor and the precise placement of treatment equipment are also constantly improving.These help to improve the cure rate of patients and improve the quality of life of patients after cure.EPID placement is a commonly used image-guided radiotherapy in China,which is of great significance to radiotherapy placement.Firstly,this paper introduces the application of radiotherapy registration,image registration and artificial intelligence in image conversion,then proposes the idea of synthesizing kilovolt DRR(Digitally Reconstructed Radiography)image from megavolt DR(Digitally Radiography)image,which is implemented by depth convolution neural network,and finally uses mutual information as a measure for registration.The research contents of this paper are as follows:(1)Introduce the application of image registration in radiotherapy.Briefly describe the basic idea of image registration,the method of image matching and the selection of optimization algorithm,and the application of artificial intelligence in radiotherapy.(2)A novel fractal hourglass convolution network for mapping DR images and DRR images is proposed.The network is based on the hourglass convolution network and the fractal theory.The network is similar to the hourglass shape at multi-scale.This morphology can make full use of multi-scale information fusion,reduce the model parameters and reduce the difficulty of training on the premise of maintaining high resolution,so as to synthesize high-precision DRR images.Pre-training the model on the soft tissue DRR generated by CT makes the model more effective and predicts the quality of DRR images better.(3)Mutual information and DR-DRR image registration algorithm based on mutual information are introduced.Different from traditional method,it used predicted DRR images of KV and the original DRR image.Registration with the original DRR image can solve the problem of registration failure caused by modal differences,and can reduce the difficulty of registration.The above method was used to analyze the actual cases in different parts.The results show that the method is acceptable.