Research Of Brain Image Segmentation Based On Multiple Features And Attention Mechanism

Glioma is one of the primary brain tumors that can produce serious consequences.It is caused by the abnormal proliferation of glial cells in the brain and spinal cord.It has the characteristics of large number of patients,high recurrence rate and difficult to cure.If early detection and early intervention,it can significantly improve the quality of life of patients.Magnetic resonance imaging(MRI)is a widely used diagnostic technique for glioma.Accurate segmentation of multiple glioma from MRI data is the premise of many accurate medical procedures.However,due to the large amount of MRI data,doctors or radiologists can’t segment the images manually in a short time,and the segmented parts can’t achieve accurate and repeated segmentation,segmentation depends heavily on professional experts.If segmentation can be performed automatically on the computer,all related problems can be solved.Therefore,the design of automatic and reliable segmentation method has become an urgent problem in the diagnosis and treatment of glioma.The existing automatic segmentation methods of brain glioma MRI are mainly divided into two kinds,one is the traditional segmentation method,the other is the deep learning segmentation method.The traditional segmentation method needs human intervention,and does not achieve complete automatic segmentation,so the robustness is poor,and the segmentation result still has a large error;the segmentation method based on deep learning overcomes the shortcomings of traditional segmentation method,does not need human intervention,and can achieve full automatic segmentation,with good robustness,and the segmentation accuracy has been greatly improved.However,the segmentation method based on deep learning still has some shortcomings,such as poor accuracy and more computing resources.In view of the above problems and background,this paper has completed the following work:1.Proposed an image segmentation model using gating and adaptive attention.According to the attention mechanism of human brain,the adaptive attention unit and gating residual unit were introduced.The attention mechanism could be used to distinguish the key regions of the image features that need attention,so as to allocate more resources to the key regions,which improves the performance of the segmentation model.2.Proposed a network model with dense channels.By increasing the number of feature extraction,the performance of segmentation model was improved,but a lot of computing resources and storage resources were consumed.3.Proposed a multiple feature extraction network model,through which more abundant features can be extracted,the computing and storage resources consumed by the network can be significantly reduced,and the performance of the segmentation model can be improved.At the same time,according to the above segmentation model,a prototype system of glioma automatic segmentation was designed.4.Proposed an image segmentation model using multi-path feature mining unit as the basic unit of network.Its unique multi-path feature mining unit could extract more diverse features according to the combination of operations.In order to use the segmentation model on edge computing devices,a lightweight multi-path mesh segmentation model was designed,which needs much less computing and storage resources than the benchmark model.