Study On The Microvascular Feature Enhancement And Segmentation Algorithm

Digital analysis of medical image is an important application of computer image processing technology in medical diagnosis.Because the morphology and structure of microvasculature are closely related to ocular fundus diseases and cardiovascular diseases: the early symptoms of hypertension,arteriosclerosis and other diseases will lead to the changes of physical properties of microvasculature network.Therefore,the accurate measurement and quantitative analysis of the whole structure,diameter,curvature and other related parameters of the microvascular network by using medical image processing technology can provide an important basis for the early diagnosis of the disease and the judgment of the disease degree.Medical image processing technology has an important significance in modern medical diagnosis.In this paper,the microvascular network structure in retinal image is studied from the aspects of feature enhancement,microvascular segmentation,angle measurement between artery and vein.The enhancement,segmentation and measurement of microvascular are realized by using Curvelet theory,depth learning algorithm,Hessian matrix and other methods.The main contents of this paper are as follows:(1)In this paper,the distribution of discrete curvelet coefficient in image is studied,and a microvascular enhancement algorithm based on curvelet theory is proposed: firstly,the green channel of image is extracted,and the pixel value in the green channel is equalized by using the CLAHE algorithm;secondly,the Curvelet coefficient decomposition of the equalized image is carried out,and a non-linear combination of continuity of soft threshold function and gradualness of hard threshold function is proposed Threshold function,according to the different distribution characteristics of noise frequency-domain coefficients and feature frequency-domain coefficients in the curvelet transform domain,realizes threshold denoising and feature enhancement.Finally,the filter algorithm based on 2D Gaussian kernel is used to filter the enhanced image,and the fusion rules of high-scale and low-scale coefficients are followed to fuse the curvelet coefficients on high-scale and low-scale respectively,so as to obtain more intuitive micro Vascular network structure.(2)This paper studies the application of deep learning algorithm in the task of medical image segmentation,and proposes an improved U-net segmentation model:first,adjust the contrast of the input samples and cut them randomly without overlapping,expand the magnitude of the samples and improve the generalizationability of the U-net model;secondly,introduce the specific step convolution and residual modules,respectively,to improve the structure of the pooling layer and the convolution layer to get more Finally,the cross loss function in the original network is replaced to make the network more suitable for the task of microvascular segmentation.The experimental results show that the improved U-net network is effective in image clarity,prevention of blood vessel adhesion and capillary segmentation.(3)Based on the above enhanced image and segmentation results,a new microvascular measurement algorithm based on Gaussian function model is proposed.Firstly,the blood vessel center point is found according to the distribution characteristics of the pixel value of the microvascular cross-section,and the measurement axis in the cross-section direction is generated by the excellent tracking characteristics of Hessian matrix.Secondly,the inflexion point of the measurement axis pixel curve is defined as the blood vessel boundary,and the Finally,the mathematical model was established to determine the relationship between the bifurcation angle and the diameter of the arteriovenous.