Automated Segmentation Of Choroidal Vessels In Optical Coherence Tomography Images Of The Posterior Eye

Choroid is a tissue which lies between the retina and sclera,surrounding the rear of the whole eyeball.It is composed of rich blood vessels and pigment cells,which can provide enough nutrition to outer retina and vitreous body.Changes of choroidal vessels make a major contribution of changes in choroid,which will cause a lot of ophthalmic diseases,such as age related macular degeneration,glaucoma,diabetic and central serous choroid-retinopathy and so on,Therefore,both segmentation and the qualitative and quantitative research of choroidal vessels are of great significance to related eye diseases prevention and treatment.In this paper,we proposed an automated method to accomplish the segmentation of choroidal vessels.Through contrast experiments,including methods contrast experiment and data contrast experiment,we can prove the reliability and robust of our proposed method.There are four parts in our framework,the pre-processing;coarse segmentation based on fuzzy connectedness;fine segmentation based on region growing and post-processing,the following is detailed introduction:i)Pre-processing:a graph-search method is applied in the raw 3D wide-view OCT(Optical Coherence Tomography)image to get both retinal layers and choroidal boundaries.And then,we can get volume of interest(VOI)by flattening the image according to the Retinal Pigment Epithelium(RPE,Retinal Pigment Epithelium).At last,a linear enhancement method is used to improve the contrast of the VOI and a 3D anisotropy filter is used to improve the signal to noise ratio(SNR),ii)Coarse segmentation,an adaptive threshold for initial seeds firstly,and then fuzzy connectedness is applied to get coarse segmentation,iii)fine segmentation,region growing is used for automated choroidal vessels segmentation.iii)Post-processing:irn this part,we remove the influence of optic disk with hessian array and adaptive threshold method.The dataset include 26 wide-view OCT images from 13 normal testers.For quantitative analysis,on one hand,we use the following four criterions to analyze segmentation result between the proposed method and ground truth.They are true positive volume fraction(TPVF,90.22%),false positive volume fraction(FPVF,0.44%),dice coefficient(DSC,82.90%)and accuracy(ACC,99.07%).On the other hand,we use average error to evaluate the repeatability,which is 10.7%.In this paper,regression analysis and consistency analysis are used,and the correlation coefficient R2 = 80.20%.The segmentation and quantitative analysis of choroidal vessels can provide more accurate information for clinicians and further help doctors make timely diagnose and treatment about related ophthalmic diseases.