Research On Image Segmentation Algorithm Of Fundus Blood Vessels Based On Deep Learning

In recent years,the segmentation of blood vessels in retinal images has always been a research hotspot in the medical field.Precise segmentation of retinal blood vessels is an important prerequisite for the diagnosis of many diseases,and is often used as an important means for diagnosing retinal vascular disease,diabetes,hypertension,and glaucoma.Traditional fundus blood vessel segmentation is done manually by doctors,but there are problems of time-consuming and excessive reliance on doctors' professionalism.With the rapid development of image processing,automatic retinal segmentation has made some progress.However,fundus blood vessel images have the problems of small data sets,different blood vessel sizes and interference of the diseased background,which increase the difficulty of image segmentation,and also cause the vascular segmentation of retinal images to be full of challenges.In summary,there is still much room for improvement in the precise segmentation of retinal blood vessels.This article selects the public CHASE data set and DRIVE data set for experiments.Read a lot of literature,choose U-Net network as the baseline model,and improve on this basis.Compared with other existing methods,the method in this paper has achieved a good segmentation effect.The main contributions of this article are as follows:1.Propose a U-Net network segmentation algorithm based on high-and low-dimensional feature fusion attention mechanism.Aiming at the problem of low retinal image segmentation accuracy,this paper embeds the high-and low-dimensional feature fusion attention mechanism in the convolution operation of the decoder based on the U-Net network.Among them,the high-dimensional and low-dimensional features contain rich categories.Information and location information,weighting the target features that need attention,reducing the interference of useless information,Improve the accuracy of the segmentation model.2.Propose a U-Net network segmentation algorithm based on series pyramid pooling module and cascaded hole convolution module.This article replaces the bottom convolution kernel of the U-Net network with a series pyramid pooling module and a cascaded hole convolution module.Both the pyramid pooling module and the cascaded hole convolution module have expanded features without increasing network parameters.The receptive field of the picture,the advantages of extracting rich image features,and the series connection method can get a better segmentation effect.3.This paper combines the attention mechanism model,the series pyramid pooling and the cascaded hole convolution module to conduct experiments.Finally,the experimental research based on the CHASE data set and the DRIVE data set shows that the proposed fusion model is combined with a single model and other phenomena.Compared with methods,the accuracy can reach 98.11% and 97.25%,and the sensitivity can reach 81.73% and 80.98% respectively.This fully verifies that the model can effectively improve the accuracy of blood vessel segmentation and assist in the diagnosis of vascular disease,which is very important for clinical medicine.Practice value.