Design And Implementation Of Interactive Medical Image Segmentation Platform

With the rapid development of artificial intelligence,computer aided medical diagnosis becomes the most common way to combine medical treatment with intelligent technology.In the process of computer-aided diagnosis and treatment,medical image segmentation is the most critical step.Doctors need to label pixel by pixel to obtain the segmentation results with existed applications.However,manual labeling is not only highly professional,but also time-consuming and tedious(especially for 3D images).With the continuous development of artificial intelligence,automatic segmentation algorithm based on deep learning greatly improves the efficiency of medical image segmentation.However,current automatic segmentation algorithms have the problems of low accuracy and insufficient robustness.Interactive segmentation algorithm optimizes segmentation results by introducing a small number of user hints,reducing labor costs and improving segmentation accuracy,which has great prospect in medical image segmentation field.However,the lack of platforms to apply such algorithms makes it difficult to implement interactive segmentation algorithms in the medical field.In addition,the interactive medical image segmentation strategy that uses 3D medical images as the segmentation target still relies on a large number of interactive operations by doctors.Therefore,an interactive medical image segmentation platform with interactive recommendation function has become a research field that needs development urgently.In response to the above problems,the main work of this paper includes:1.Designed and implemented an interactive medical image segmentation platform: In this paper,an interactive medical image segmentation platform is designed and implemented to liberate doctors from the complicated and purely manual labeling,a front and back end separation architecture mode is adopted.The platform contains three subsystems: a React-based medical image display subsystem,an interactive algorithm-based medical image segmentation subsystem,and a deep learning-based interactive recommendation subsystem.2.Designed and implemented a medical image display subsystem based on React: This subsystem mainly provides the function of loading files and displaying images.Basically,it extends the advanced functions of viewing medical images such as section switching,image adjustment,three-view display and multiwindow display,laying a foundation for the subsequent interaction between doctors and algorithms.3.Designed and implemented a medical image segmentation subsystem based on an interactive algorithm: This subsystem provides an environment for interactive medical image segmentation,aiming at freeing doctors from heavy workload of pure manual labeling and poor robustness of pure automatic segmentation.Doctors can obtain accurate segmentation results only by simple error correction interaction,which greatly improves the efficiency of doctors' segmentation.Meanwhile,it persists the error correction information and then regularly trains and replaces the model to further improve the segmentation accuracy.4.Designed and implemented the interactive range recommendation subsystem based on deep learning: Due to 3D medical image relying on a large number of interactions in the process of interactive segmentation,this subsystem provides the recommendation function of interactive range,mainly from two aspects of slice recommendation and region recommendation.For slice recommendation function,this paper uses heuristic algorithm to select the most representative slice,so as to further reduce the influence of the large number of 3D image slices on the segmentation efficiency of doctors.For region recommendation function,this paper trains a confidence network and selects the region with low confidence based on its inference results,so as to help doctors identify the region ignored by the segmentation algorithm accurately,quickly and efficiently.All the above work has been verified in this paper.For the system implementation,this paper has carried on the full function test and compatibility test? For the interactive range recommendation function,the effectiveness of the recommendation is proved by various experiments.