Research On Brain Tissue Segmentation Methods Of MRI

Magnetic resonance imaging(MRI)technique has been widely used in radiology to visualize internal structures of the body in detail.It has some notable advantages like high spatial resolution,non interventional,good discrimination of soft tissues,flexible scanning angle,variation of scanning parameters,and harmless to the patient,etc.It has become an important auxiliary means for the diagnosis of brain diseases.It is of great significance to accurately segment the brain tissue for the analysis and research of the subsequent dissection of brain diseases,such as Alzheimer’s disease,multiple sclerosis,Parkinson and schizophrenia.Due to the distinctive physical properties of the brain tissue,the MR images generally show different grayscale intensity range,and the Gaussian Mixture Model(GMM)has become an ideal model to describe the slow change of gray level.However,the traditional GMM is based on the assumption of the independence of pixels,and the spatial structure information is often ignored.At the same time,due to the complex structure of the brain tissue,coupled with the imaging process,such as the bias field also called intensity inhomgenelty,partial volume effect(PVE),noise and other physical factors,resulting in prewise constant properties of ideal tissue intensity distribution are violated.In order to improve the accuracy of brain tissue segmentation,this paper focuses on the research of 3D brain tissue segmentation based on Gaussian Mixture model based on prior information and posterior neighborhood information.Specific research contents are as follows:Firstly,this paper presents a new segmentation algorithim(PA-GMM)which is based on the traditional Gaussian Mixture Model and the prior anatomical structure information of brain tissue probability maps.The experimental results show that the PA-GMM algorithm can solve the problem of the traditional GMM because of the lack of spatial information,which leads to the increase of the error rate in the presence of noise and bias field.Secondly,there is a bias field in the MR image,and when the bias field is too large,it will seriously affect the accuracy of the final algorithm segmentation.Therefore,on the basis of PA-GMM,this paper implements a joint segmentation framework for simultaneous calibration of bias field correction and segmentation,which extends thePA-GMM well.Experimental results show that the proposed method can quickly and effectively segment MR images in 3D.Compared with the traditional preprocessing stage,the bias field correction is performed by the alternating iteration of the segmentation and the bias field.Thirdly,in order to further improve the algorithm’s segmentation accuracy when there is a high level of noise.We utilize the spatial anatomical structure of the prior information of the brain tissue probability maps and posterior neighborhood information to redesign the expression way of mixing coefficients,proposes a segmentation algorithm which is called SNPA-MGMM.The algorithm not only can be used to suppress noise,but also can preserve the complex overlapping regions such as GM and CSF and the detail information of edges.Finally,we choose the Brainweb’s simulated data sets and the IBSR’s real data sets(IBSRv1.0 and IBSRv2.0)as our experimental data.Do experiments with improved algorithm and unimproved algorithm,and compare the results with some recent literature and segmentation results on medical software.At last,we using expert manual segmentation(commonly known as the gold standard)for quantitative analysis.Experimental results show that the proposed segmentation approach improve the results.