Research And Development Of Tumor Medical Image Registration System In Proton Radiotherapy

Cancer is one of the major diseases seriously endangering human life in the world,and the research on cancer treatment is also an important part in the field of medical research.Radiotherapy uses energy rays to irradiate the lesion area to kill cancer cells.Because it can achieve local treatment and has good efficacy,it is an irreplaceable modern cancer treatment method.The sub-project "Image Guidance and Treatment Planning System" of "Research and Development of Proton Radiotherapy Equipment Based on Superconducting Cyclotron" puts forward the need to determine tumor deviations.This paper presents a method to determine the patient’s positioning error during radiotherapy and further determine the tumor offset.The main contents are as follows:1.A segmentation network is designed to segment tumors in medical images with high precision,and a tumor segmentation method based on deep learning is presented.The processing flow of data set processing,model design and training,and image postprocessing is introduced.The network model used is the ResU-Net network structure designed by combining the residual structure and the advantages of U-Net,and the ability of ResU-Net and traditional U-Net to segment non-small cell lung tumors is compared through experiments.By using the ResU-Net network model based on the residual structure and the Dice loss function,the segmentation accuracy is higher.After adjusting the parameters,the segmentation Dice coefficient of the optimal model is 0.8676±0.1537,the sensitivity is 0.8550±0.2081,and the specificity is 0.9997±0.0004.2.A registration algorithm that can register multi-modal images is given,and the algorithm is tested and tuned using aligned simulated multi-modal images.The registerable range of the algorithm is also tested.Finally,it is proved This algorithm is suitable for registration between CT and CBCT images.The method of artificially moving the tumor position in the image is given,and an experiment is designed to test whether the registration method can obtain the offset when the tumor moves relative to the human body.After testing and analysis,a registration method that can obtain the tumor offset is given,that is,use the tumor segmentation results to assign the non-tumor area to-1000,then use the cuboid ROI containing the tumor to cut out part of the image block,and finally register the two ROI image block.The translational deviation of the tumor offset obtained by this method is less than 1mm,the angular deviation is less than 1°,and the time-consuming is less than 1 minute.3.The process of tumor medical image registration is encapsulated into the software of "Tumor Medical Image Registration System".The user interface is designed.The design idea,function calling logic and technical configuration of the software are introduced.The medical image tumor registration system is realized,and the basic operation process of the software is displayed.Finally,the ability of the software to measure the deviation of the tumor is tested by using the registration method between the CT image with tumor offset and the original CT image.The registration process is stable,the registration speed is within one minute,the registration translation deviation is less than 1mm,and the angle deviation is less than 1°.