Interactive Medical Image Segmentation For Multimodal MRI

With the rapid development of computer technology,medical image segmentation has gradually become an important basis for assisting doctors in making diagnostic decisions.However,segmentation of medical images by manual segmentation will consume a lot of labor costs and bring pressure to the doctor’s life.In recent years,in the context of the continuous development of artificial intelligence technology,a large number of semantic segmentation algorithms based on deep learning have emerged in the field of medical image segmentation.These algorithms have achieved certain results in segmentation accuracy and segmentation efficiency.However,most of the current medical image segmentation methods do not effectively use multi-modal medical image data,and do not pay attention to the correlation between tumor data tags,which makes a large amount of multi-modal medical image data not effectively used.In addition,the current semantic segmentation algorithm is not enough to be directly applied to clinical use,and it requires interactive adjustment by doctors.The existing interactive segmentation methods can only deal with singlemode and single-target tasks,so the problem of multi-modal interactive medical image segmentation has become a difficult problem.Otherwise,current medical image segmentation methods often ignore a large amount of image detail information due to downsampling operations on 3D image data.How to solve this problem has become another important issue in current medical image segmentation algorithms.In addition to algorithmic defects,the existing medical image processing platforms also have certain problems in actual use.The segmentation model integrated in the platform is often difficult to generalize to the data of a specific hospital.Moreover,most medical image processing platforms only integrate traditional image segmentation algorithms,which makes it difficult to follow academic development.In response to the above problems,the work of this paper is as follows:1.In this paper,the multi-classification problem in the medical image segmentation task is modeled as a hierarchical multi-label prediction task,and on this basis,an interactive medical image segmentation algorithm for multi-modal data is proposed.The algorithm proposed in this paper fully considers the structural commonality between multi-modal data.Moreover,this paper introduces multimodal data aggregation and multi-objective hierarchical output structure on the existing interactive image segmentation model.Finally,this paper has conducted a lot of experimental analysis on the algorithm.The experimental results show that the algorithm proposed in this paper can effectively reduce the number of user interactions while ensuring the segmentation effect.2.This paper proposes an automatic medical image segmentation refinement algorithm based on global-local.The global-local idea is used to build a local refinement module,and on this basis,the existing segmentation results are optimized.Then,this paper uses reinforcement learning to model the dynamics of the refinement process,so that the model can automatically focus on the areas worthy of local refinement based on the existing coarse segmentation results.The method of reinforcement learning is used to simulate the actual refinement process of the doctor,which effectively improves the efficiency of local refinement and the segmentation effect.3.This paper designs and develops a new medical image segmentation processing platform system,and integrates the above two algorithms into the platform.In addition,the platform designed in this paper also implements basic medical image display and operation functions.More importantly,the platform designed in this paper provides users with a semantic segmentation function based on deep learning.On this basis,this platform supports users to retrain models based on their own data,and also provides academic researchers with a unified interface to facilitate the deployment of the latest research work to the platform.