Multiple Sclerosis Lesions Segmentation Method Research Based On Multi-Atlas And Multi-Channel

With the continuous application and development of artificial intelligence and computer technology in medical image processing,digital medical image has laid a foundation for clinical diagnosis and auxiliary diagnosis.The processing and analysis of brain images have attracted extensive attention from experts.In digital medical image processing,image segmentation is a key preprocessing step in image recognition and computer visualization.The purpose of digital medical image segmentation is to segment and study regions of Interest(ROI)in images.In addition,efficient and accurate segmentation is an important step in quantitative and qualitative analysis and 3D visualization.Magnetic resonance imaging is often used to describe and quantify multiple sclerosis(MS)in the brain and spinal cord lesions.The number and volume of lesions have been used to assess the disease burden of multiple white matter lesions in the brain,track disease progression,and evaluate the effectiveness of new drugs in clinical trials.It is extremely difficult to accurately identify white matter lesions in MRI images because of differences in location,size,shape,and anatomy between subjects.Based on the traditional Markov random field theory,this paper proposes an improved multiple sclerosis lesion segmentation method based on multiple-atlas and multiple-channels according to the characteristics of white matter lesions.The research work is as follows:(1)Literature review and analysis.Firstly,a large number of literatures related to brain MRI imaging and medical image segmentation were reviewed to systematically analyze the main research status and future development trend of domestic and foreign experts in the field of medical image segmentation.Secondly,the basic principles of MR imaging technology and digital medical image segmentation are described.Finally,combining with the characteristics of white matter lesions and the advantages of existing segmentation techniques,this paper proposes the segmentation method of white matter lesions.(2)Selection of image preprocessing methods.The data sets used in the experiments were publicly available from MRI scans and had not been otherwise processed.In order to makesegmentation more accurate,save time and improve robustness,gray level irregularity and deviation field correction,decertification and other operations were carried out.(3)The Bayesian theory,Markov random field and Gibbs random field principle are systematically described,and their applications in medical image segmentation are analyzed.In addition,the shortcomings of traditional medical image segmentation methods are described.On this basis,we use the energy minimization method to correct and segment the migration field.Finally,T1-w,T2-w and FLAIR images of the same patient were labeled and fused to make the final segmentation more accurate.(4)Experimental verification and evaluation.In this paper,Matlab2015 b,FSL and MIPAV software platforms were used to verify and analyze the feasibility of the proposed segmentation method,and to evaluate the segmentation results quantitatively and qualitatively.The innovation of the research is that(1)The adaptive sparse bayesian decision theorem was used for MS Lesions segmentation,which was combined with MRF and GRF to adopt a reliable and robust Multiple sclerosis lesion segmentation method.The purpose is to improve the segmentation accuracy and reduce the processing time;(2)Multi-channel mechanism combining T1-w,T2-w and FLAIR images are used to perform tag fusion,with the purpose of reducing matching ambiguity and improving segmentation accuracy;(3)The effective calculation of the size and number of lesions can help experts to carry out prognosis and analysis.The deficiency of the study is that the number of iterations for correction of the gray irregularity of the data is relatively large,so the overall treatment efficiency of white matter lesions does not reach the desired effect.The accurate calculation of the volume and number of lesions will facilitate the diagnosis and analysis by doctors,but our method still fails to achieve the desired effect.