Medical Tissue Image Segmentation Based On Level Set Method

The diseases pertaining to retinal could change retinal vessels?organizational form and structure leading to visual degeneration or loss.For most ophthalmic diseases,doctors can make early discover?prevention and cure by testing and analyzing retinal images to decrease the rate of losing vision.However,no matter 2D colorful fondues or 3D optical coherence tomography,it will be very time-consuming as well as disturbed by subjective consciousness to make quantitative and qualitative analysis for visual disease under manual segment for the sake of big data set.For the causes above,it is urgent to have some ways to segment automatically to help doctors judge the visual diseases and make precise analysis and cure.Retinal vessel image has such properties as complicated net structure,big range of dimension,low contrast ratio between microvascular and background,focus and noise,which lead existed vessel segment methods cannot adequately segment adjacent vessels?cross vessels and microvascular.Aiming at the shortcomings of existed methods,this paper proposed level set vessel segment method which combined regional energy fitting information and shape prior.The method firstly uses morphological operator and Gaussian convolution to get reinforced retinal vessels images;then analyses the differences of eigenvalue of Hessian matrix on vessels and background,reconstruct a vessel response function to primarily evaluate the vessel image as shape prior and initialization information of level set model and build a level set vessel segment model including regional energy fitting and shape restrict;finally,use the area?width and height of connected domain to construct geometric factor to further eliminate fake shadow and focus of smaller connected domain area to get final result of vessel segment.Technology of 3D optical coherence tomography can harmlessly get high resolution tissue image reflecting clearly the formation and structure of retinal organization.However,the influences of macular center,diseases and imaging artifact cause great challenge to segment retinal layer segmentation.For the reasons described above,this paper proposed a full automatic retinal layer segmentation method based on elastic energy fitting.The method primarily alleviated the interfere of speckle using anisotropic diffusion model to filter primitive retinal images;then used Canny test operator to extract the area of interest and linked upper borders of cilia;finally,used multiple level set model including regional energy fitting,area restrict in layers to complete retinal segment on ten different layers.The experiments of methods proposed in this paper were executed on there 2D colorful fundus image databases HRF,STARE,DRIVE,respectively,sensitivity are respectively up to 79.4547%?79.0860%?75.3535%,accuracy are respectively up to 96.1820%?95.0340%?95.3565%.The data used to experiment come from 3D-OCT1000(Topcon Corporation,Tokyo,Japan)who got the scanning image of macular center from 15 people,the deviation of average unsigned border location is 5.90±4.57?m.