An Adaptive Sparse Bayesian Model Combined With Multiple-Atlas Label Fusion For Tumor Segmentation In Brain MRI

In recent years,medical image processing technology develops rapidly,and brain image processing is a very important part of it.Image segmentation is to segment and extract the region of interest,which is a very important for the subsequent surgery planning or other works.MRI is one of the important approaches in brain tumor diagnosis and the processing of MRI with brain tumor is one of the necessary steps for clinic.Tumor segmentation on MRI has got more and more attention.This paper focus on the tumor segmentation in brain MRI using an adaptive sparse Bayesian model combined with multiple-atlas label fusion methods,which not only segments the region of brain tumor,but also segments the subregions of tumor.The main work are as follows:(1)Relevant literature research and analysis.The literatures related to brain tumorsegmentation at home and abroad were investigated.The present research situation andfuture development trend of medical image segmentation are analyzed.(2)Introduction of the theory of image segmentation.The sparse Bayesian model andMarkov field model are introduced systematically.(3)Introduce the automatic segmentation of brain tumor on MRI using the sparseBayesian model and the fusion of different modality by multiple-atlas label fusion.Thefirst focus is an algorithm that applies the sparse Bayesian decision theorem combinedwith the Markov random field and Gibbs random field for the segmentation of braintumor in T1-w,T1-c,T2-w and FLAIR image,respectively.The second focus is themulti-atlas-based label fusion algorithm using local weighted voting strategy,in whicha specific coefficient as the weight is assigned to each individual atlas.The coefficientis calculated using the maximum likelihood function between atlas and the targetimage.(4)In the experiment and verification section.The proposed approach is evaluated bysome training cases obtained from MICCAI Brain Tumor Segmentation Challenge2017.The results were analyzed qualitatively and quantitatively.A comparison of theproposed approach and other approaches is presented in terms of the true positive rate(TPR)and the positive predictive value(PPV).Furthermore,the total tumor volume iscalculated.It can be seen that the proposed approach is able to acquire a larger meanvalue of the Dice similarity coefficient(DSC)than the other approaches do.At the end of the paper,make an overview of the whole paper and what I did.Making an analysis of the advantages and disadvantages of the research method.Besides,the future planning is also proposed.