Research And Application Of Multiple Sclerosis Lesion Segmentation Based On Multi-atlas

Multiple sclerosis is a kind of common diseases of the central nervous system,early symptoms include numbness and weakness of limbs,later can lead to stroke,cognitive impairment and vision degradation and so on,serious threaten human’s health.Clinically diagnosed the illness by sketching pathological tissues according to MRI,which time-consuming and subjective uncertainty.Therefore,researching automatic segmentation algorithm of multiple sclerosis lesions and improving the accuracy and stability of segmentation,is of great significance for the diagnosis and treatment of the disease.On the basis of reading a large number of literature,this paper proposes a multiple map registration method for the automatic segmentation of multiple sclerosis lesions.Research work mainly include:(1)Introduction.This part introduces the research significance of multiple sclerosis lesions segmentation,analyzes the present domestic and foreign research status about multiple sclerosis lesions segmentation by reading the literature,makes a brief description to all parts of this article.(2)Magnetic resonance image segmentation method.Brain lesions of necessary theory knowledge is introduced,including the related knowledge of brain medical image,magnetic resonance imaging and the method of magnetic resonance image segmentation.Make a brifly introduction for the threshold method,the clustering method,watershed algorithm and wavelet transform.There are three kinds of segmentation method based on graph: segmentation method based on a single map,based on the average map and based on multiple map.(3)Multiple sclerosis lesions image segmentation algorithm based on map.Make a introduction for segmentation method based on multiple mapping process,and the process of image preprocessing,registration process and image fusion has just sent the introduction of the three steps in detail.In the preprocessing process,first skull stripping using BET algorithm,then using the curve fitting and c-means fuzzy clustering method for deviation correction.In terms of registration,this article mainly aims at the problem of low registration accuracy,usingthe FSL-anat registration method.Fsl-anat registration method using local similarity as the similarity measure of registration,and the strain field on the effective constraint and smooth can achieve a good result for registration.We use FSL-anat registration as a fusion of local similarity weights,similarity measure high pixels,illustrate a moe accurate registration,the greater the effect on the final segmentation result.The weighted selection fusion strategy is used in the fusion process,and according to the characteristic of brain magnetic resonance images,puts forward the algorithm of this paper.(4)Experimental results analysis of Multiple Sclerosis Lesion segmentation based on Multi-atlas.Using the algorithm proposed in this paper to experiment and implement the segmentation of multiple sclerosis.According to the characteristics of the brain lesions,this article selects the similarity measure to evaluate the segmentation result.Fusion of brain lesion is similar to result of expert manual segmentation whether its shape or size..All 10 groups of experiments have reached 0.8 of dice similarity.It proves that Multiple Sclerosis Lesion segmentation based on Multi-atlas can reach a high segmentation accuracy of lesions.