MS Lesion Segmentation Algorithm Research Based On Fuzzy C-means And Probabilistic Label Fusion

Brain Magnetic Resonance Imaging(MRI)technology has been widely concerned by researchers and radiology doctors due to its non-invasiveness,high processing speed.high resolution,and non-ionizing damage to subjects.It is an important technical means for clinical diagnosis and research.Brain magnetic resonance image segmentation is an important step in computer-aided medical diagnosis.It provides the shape,location and statistical information about the brain's healthy tissue structure and the lesion area.It not only assists in the quantitative study of brain lesions.but also can dynamically diagnose and analyze changes in brain tissue structure and the development process of lesions.It plays a vital role in the prediction,treatment and prognostic diagnosis of brain diseasesThe pathological features of multiple sclerosis(MS lesion)are demyelination,shedding of axons,and increased glial cells.The correlation between the onset of lesions and age shows that the younger the age of the disease,the more severe the multiple sclerosis lesions and brain atrophy Multiple sclerosis lesions are mostly located around the ventricle,especially in the subventricular veins near the ventricle body and the lateral ventricle corners.Lesions show different brightness signals in different modes of Magnetic Resonance(MR)images,medium signals at Tl-w,high signals at T2-w,and high signals at FLAIR.Clinical experts usually draw the lesion boundaries manually on many brain MRI image slices based on the pathological and anatomical characteristics of MS lesions and imaging features in MRI,then consider the spatial relationship between the lesion and surrounding tissues according to the depicted boundaries to determine the treatment plan of the patient.However,manual segmentation by imaging experts is time-consuming,cumbersome,and susceptible to the expertise and experience of the experts themselves.Therefore,how to effectively segment MS lesions in MRI images has been an important research topic for experts and scholars in recent yearsThis paper proposes an MS lesion segmentation algorithm based on fuzzy C-means and probabilistic label fusion.This algorithm can effectively segment MS lesions in MR images.Select 20 public data sets of the MICCAI 2008 MS lesion segmentation challenge.Each MR image in the data set contains three modes,T1-w,T2-w and FLAIR.The isotropic resolution of each mode is 0.5mm and the size is 512*512*512.The main steps of the algorithm are:(1)In this paper,Medical Image Processing,Analysis and Visualization(MIPAV)software is used to preprocess the data set.The preprocessing operations include downsampling,skull-stripping,and normalization.Finally,an MR image with a size of 256*256*256,only containing the intracranial space of the brain and reduced gray deviation is obtained;(2)The improved SLIC superpixel algorithm is used in the MR image segmentation preprocessing step to obtain MR images containing K superpixels This algorithm can effectively suppress the impact of noise on MR image segmentation.Superpixels retain effective information for image segmentation.Using superpixels instead of pixels as the basic unit of image processing can reduce the complexity of data processing and increase the accuracy of image segmentation;(3)Superpixel-based fuzzy C-means algorithm is used to segment MR images with different modal MS lesions to solve the problems of uneven grayscale and blurred borders of MR images.The segmentation Accuracy of multiple sclerosis is improved by introducing the neighborhood information constraint of superpixel;(4)The use of probabilistic label fusion algorithm in the multi-modal MR image fusion step can dynamically combine the segmentation advantages of each modal.This algorithm is based on a locally weighted voting strategy,using different model segmented results of the same patient that have been registered as input.The maximum likelihood function between the target image and a given modal image that has been calculated is used as the unique weight coefficient of each modal,and the final MS lesion segmentation result is output after fusion.(5)By comparing the results of automatic segmentation and expert segmentation,the evaluation indexes DSC(Dice Similarity Coefficient),PPV(Positive Predictive Value)and TPR(True Positive Rate)of the algorithm in this paper are obtained.The segmentation results of our algorithm are better than other segmentation algorithms.which proves that our algorithm has significant effect for improving the segmentation accuracy of lesions in MR images.