Research On Automatic Segmentation Of Retinal Vessels Based On Optimized Multiscale Linear Operator

Fundus image analysis plays an important role in medical diagnosis,and retinal blood vessels are the only observable blood vessels under non-invasive conditions.Some physiological structural changes of blood vessels can characterize many diseases.With the development of computer technology and artificial intelligence,computer-aided ophthalmologists can improve the efficiency of disease screening by treating fundus images.Many automatic retinal vessel segmentation methods have been proposed in the existing literature,but most of the methods have the problem of reducing the accuracy of vessel segmentation,such as the intermediate cavity,the partial small blood vessels and the background noise mixed and lost in the bloo d vessel segmentation,and at the optic nerve head boundary.Pseudo blood vessels will be produced.Therefore,this paper proposes a method based on genetic algorithm to optimize the combination of multi-scale linear operator and optic nerve head filling,to solve the problem of large blood vessel intermediate cavity,reduce background noise and pseudo-vessels.(1)This paper uses genetic algorithm to optimize multi-scale fusion coefficients.Firstly,a morphological opening operation is performed on the green channel of the retinal blood vessel image using a circular structure element of 1 pixel width to process the central light reflection of the blood vessel,and then Gaussian noise filtering is performed on the image by using a Gaussian kernel function filter with a scale of 9 and a standard deviation of 1.8.Then the background subtraction method is used to obtain the gray balance image of the image.Then,the linear operators of different scales are used to detect the blood vessels of different directions,scales and intensities,and the genetic algorithm is used to optimize the parameters of the multi-scale operator weights.For the best detection accuracy.(2)In this paper,the optic nerve head filling algorithm is used to remove the pseudo blood vessels.Firstly,using the gray feature of the optic nerve head,the optic nerve head is roughly segmented by a threshold,and then the morphological expansion is used to reduce the incomplete effect of the optic nerve head segmentation.A set of pixels to be filled is obtained,and the current pixel gray value and the neighborhood pixel gray level are compared.The size of the mean,according to the comparison results,selectively fill the highlighted pixels in the optic nerve head region and preserve the vascular tissue in the optic nerve head as much as possible.The experiments in this paper are mainly carried out on the open source DRIVE and STARE databases.The evaluation criteria are divided into accuracy,sensitivity and specificity.The experimental results show that the proposed method is effective,and the DRIVE and STARE databases obtain the accuracy of blood vessel segmentation of 93.90% and 95.32%,respectively.