Research On Mri Interactive Segmentation Method Base On Deep Reinforcement Learning

Interactive segmentation is a method of adding artificial marks to the existing supervision information based on automatic segmentation,so as to implement a more refined segmentation method.This method,which can further optimize the segmentation results with a small amount of hint information,has certain research value.At present,the existing interactive segmentation methods have been applied in various fields,especially in the medical field.Doctors can get more refined segmentation results by labeling and prompting the images.However,the current interactive segmentation method needs to carry out targeted processing schemes according to the characteristics of different images.For example,on medical images,doctors need to mark the lesion area multiple times to get better semantic information,so a lot of manual interaction is still required.,which is a great burden on the staff.Therefore,we urgently need a method that can guarantee a certain accuracy and require less interactive information.Existing interactive segmentation methods use the results of automatic segmentation to replace the initial information of user interaction,and update the results interactively through multiple iterations.However,this update method does not consider the characteristics of the image itself.It only interacts according to the shape and pixels,and lacks fine processing of edge information,thus losing more detailed information of the image.Therefore,more interactions are often required to correct the obtained results..How to solve this problem becomes a difficult problem in interactive segmentation.In addition to the defects in the algorithm,the existing image processing platform system is difficult to be applied in actual engineering due to the poor generalization ability of the integrated model,and most systems only use the traditional image segmentation method,which is out of track with academic development and difficult to update iterate.1.This paper uses the idea of reinforcement learning to treat each pixel as an intelligent agent,and models the scene as an interactive process.The reward mechanism is based on user guidance,and the aim is to enable the intelligent agent to learn to make the correct segmentation action based on different image feature information.The existing interactive segmentation model is improved by introducing the attention mechanismbased edge processing module and the feature diffusion mechanism proposed in this paper.2.This paper proposes an interactive segmentation method based on attention-feature diffusion.Based on the idea of deep reinforcement learning,a channel attention module is built to strengthen the agent’s processing of edge information,which enriches semantic information and obtains a finer boundary.process result.At the same time,the diffusion algorithm is introduced to regard the user prompt feature information as a path,thereby strengthening the agent’s ability to process local area information,effectively improving the efficiency and segmentation results of local segmentation,and reducing the number of interactions.3.To facilitate the application of algorithms to practical needs,this paper develops an interactive system and adds relevant interfaces for users to use.