Segmentation Of Retinal Blood Vessels In Ocular Fundus

In clinical medicine,ophthalmologists need to refer to the segmentation status of retinal blood vessel images in the fundus to diagnose fundus diseases.Therefore,using artificial intelligence technology to accurately segment retinal blood vessel images will play a certain auxiliary role for ophthalmologists in diagnosing fundus diseases,not only reducing the pressure on ophthalmologists,but also shortening the time for patients to visit.This article has conducted research on retinal blood vessel segmentation on three publicly available datasets,DRIVE,CHASE,and STARE,and designed two different network models.Experimental results have verified that the proposed two network models can effectively and accurately segment retinal blood vessels.The content of this article includes:Firstly,in retinal blood vessel segmentation,aiming at problems such as small blood vessel breakage,excessive information loss during segmentation,loss of blood vessel endings,and the need to improve segmentation efficiency,the structure of a fully convolutional neural network(U-Net)with encoder decoder is improved and optimized,and an RSA-Unet algorithm is designed combining the advantages of residual error and spatial attention mechanism to segment retinal blood vessel images.The residual module used by the RSA-Unet method can not only be used to build deep networks,but also improve network performance and obtain deeper vascular features,effectively solving the problem of feature loss and information loss in retinal blood vessels.The spatial attention mechanism introduced by RSA-Unet can enable the network to fully learn which information is meaningful and important,that is,the spatial attention mechanism can enhance important features(such as vascular features),and can suppress unimportant features,which can effectively improve the retinal blood vessel segmentation effect.Secondly,to address the current problems of incorrect segmentation,insufficient segmentation of small blood vessels,and improved accuracy in retinal blood vessel segmentation,an improved algorithm for retinal blood vessel segmentation based on U-Net is proposed.A CAMRes DBU-Net network is designed,which combines the Channel Attention Module(CAM)and Residual Dense Block(Res DB).This network first utilizes the channel attention mechanism to enhance the recognition ability of the network,Then,the residual density module is used to replace the traditional convolution module to improve the performance of network segmentation of small blood vessels.At the same time,the residual density module of this network can replace the traditional convolution of encoding and decoding,achieve adaptive calibration of the features of retinal blood vessel images,highlight significant information and suppress irrelevant information,which can well maintain more detailed information of blood vessels,achieving accurate segmentation.The two segmentation algorithms proposed in this article have been experimentally validated on three public datasets,DRIVE,CHASE,and STARE.Through comparison with the retinal blood vessel segmentation algorithms proposed in recent years,it is found that the two algorithms proposed in this article have shown good performance in various performance indicators,achieving accurate segmentation.