Research And Application Of Fundus Retinal Vascular Segmentation Based On Convolution Neural Network

Retinal images contain many features related to cardiovascular disease,and manual segmentation of blood vessels takes a long time and is prone to errors.In this case,artificial intelligence technology is particularly important,which can improve doctors’ work efficiency.However,retinal blood vessels have different contrast,easy to be confused when segmenting the cross area of blood vessels,and the influence of lesions in some parts.In order to deal with these problems and improve segmentation performance,especially when dealing with focal areas and small vessels,the following studies were conducted in this paper:(1)Today’s division methods have problems such as low accuracy,easy to be affected by the quality of picture and lesions.To this end,this article proposes a multi-scale segmentation model that combines the multi-scale segmentation model that pays attention to the residual network,MSRA-UNET,which is used to improve the recovery effect of real images.This method is based on the U-NET network and uses the residual attention module to alleviate the problem of information loss caused by poolization and convolution.Through multi-scale input,the characteristics of different feelings can be extracted.Use the aggregation module to aggregate these different scale feature diagrams to obtain more information.Then add a parallel space activation module to the upper sampling part,which can further highlight the boundaries of the blood vessels.Through the ablation experiments on Drive,Stare,and Chase＿db1 datasets,cross-training verification experiments were conducted and experiments were compared with previous methods.(2)In order to solve the problem of insufficient feature extraction and noise and low accuracy,this article introduces the structure of attention mechanism,multi-scale input and multi-branch structure,and builds a multi-scale multi-branch of a multi-scale and multi-branch of a fusion mechanism.MSMBA-UNET is used for retinal vascular segmentation.This model can extract the images of different scale,and use the empty convolution to further enhance the feature extraction ability,which can effectively extract and integrate the feature information of different scale and branches.At the same time,the attention mechanism is added to obtain richer details Information,thereby improving the accuracy of the segmentation of the retinal vascular segmentation.Through the ablation experiments on Drive,Stare,and Chase＿db1 datasets,cross-training verification experiments were conducted and experiments were compared with previous methods.In this paper,the improved convolutional neural network model is used to segment retinal blood vessel image,which can effectively overcome some problems existing in the existing retinal blood vessel segmentation methods,and achieve more accurate segmentation effect.