Research On Method Of Multiple Sclerosis Lesion Segmentation Base On Attention Mechanisms

The development of brain science and artificial intelligence are complementary each other,which have been paid more and more attention.Brain medical imaging analysis is the intersection of brain science,medicine and informatics.It aims at analyzing a large number of brain medical imaging data efficiently and accurately,and improve efficiently of professional doctors.Recent years,with the arrival of the third upsurge of artificial intelligence represented by deep learning,neural networks based medical image analysis methods have achieved great success.Multiple sclerosis(MS)is an autoimmune disease caused by central nervous system injury,which often affects brain tissue such as white matter,optic nerve and cerebellum,etc.Magnetic resonance imaging(MRI)provides sufficient imaging contrast to visualize and detect lesions.Quantitative measurement methods based on various features of the lesion have been shown to be useful are proved to be useful for evaluating treatment methods in clinical trials.Therefore,robust and accurate segmentation of MS lesions from MRI images can provide important information about disease status and progression.The neural network model designed from the perspective of attention mechanism,context information guidance and distant dependence are used to segment MS lesions.Through multi-dimensional analysis of multimodal magnetic resonance imaging data,the automatic and accurate segmentation of MS lesions are realized.It provides the basis for quantitative analysis MS lesions,which can assist doctors in diagnosis and treatment MS.According to the characteristics of MS location,size,quantity and shape of the lesion,a three-dimensional context guidance module based on Croneck convolution is designed to integrate the focus information of different visual fields.Then 3D spatial attention module is introduced to enhance the recognition of focus features in MRI images.Focus loss function is designed for the existence of imbalance between MS and non-focal samples.The attention context U-Net for MS lesion segmentation is proposed by using depth monitoring and coding-decoding structure.To further extract the multi-dimensional MS focus information of MRI images and effectively identify the regions of MS lesions,this paper proposes a cross-dimensional cross attention mechanism and a multi-dimensional feature similarity module for MS lesion segmentation network.The cross-dimensional cross attention module skillfully integrates the capture of context information based on Kronecker convolution,the integration of spatial and channel information,the aggregation of 3D voxel information and 2D pixel information.At the same time,the multi-dimensional feature similarity module is designed to aggregate the remote dependency information of multiple dimensions.Furthermore,we develop a multidimensional cross attention U-Net,which can capture the information of MS lesions from multiple dimensions,channels and fields of vision.The experiments confirm that the context-guided module,cross-dimensional cross attention module,multi-dimensional feature similarity module designed for the characteristics of MS lesions are effective.the proposed attention-context U-Net and multi-dimensional cross-focus U-Net have great advantages in the objective evaluation index of selected lesion segmentation.The research results of this paper provide a fresh idea for MS segmentation.