Research And Application Of Retinal Vascular Segmentation Algorithm Based On Deep Learning

Research indicates that some common diseases,such as hypertension,diabetes,cardiovascular diseases,and thyroid disorders,may lead to retinal vascular changes in the eye.Currently,the clinical diagnosis method involves manual detection of the structure of retinal vessels by doctors and manual segmentation of the vessels to obtain retinal vascular information.This method is inefficient and costly,presenting challenges in both effectiveness and timeliness for ophthalmologists conducting retinal vascular structure detection.Therefore,this paper considers utilizing the relevant theories,methods,and techniques of deep learning to conduct in-depth analysis and research on retinal image segmentation and artery-vein classification,and to make it more practically valuable.To address the problems in retinal vessel segmentation,this paper proposes improving the U-Net neural network model to achieve segmentation and artery-vein vessel classification of re tinal vessels.The U-Net neural network has shown excellent performance in medical image analysis,effectively segmenting retinal vessels.This paper experimentally analyzes and compares the segmentation performance of the U-Net neural network and its derivative networks.A self-attention mechanism is used in the vessel segmentation network to ensure that the neural network pays more attention to the semantic information of vessels and reduces external information interference.The network can continuously modify the segmented vessel images through iterative training to obtain more accurate segmentation results.These two approaches are combined to achieve high-quality segmentation of retinal vessels.Artery-vein vessel classification of retinal vessels is a lso achieved based on segmentation and using the U-Net neural network.Data augmentation techniques and semi-supervised learning methods are used to address the problem of insufficient labeling data for artery-vein vessels.The U-Net neural network is used for concatenated training of retinal vessel arterie and veins,achieving artery-vein classification on the basis of vessel pixel-level segmentation,while ensuring the accuracy of artery-vein vessel classification.Finally,experiments on the DRIVE and HRF public datasets are conducted for vessel segmentation and artery-vein vessel classification.The results show that the proposed improved network achieves good segmentation performance in terms of accuracy and timeliness for retinal vessel segmentation and artery-vein vessel classification.Furthermore,based on the above research,we have constructed an intelligent system for retinal vessel segmentation and artery-vein vessel classification,which can better assist doctors in clinical analysis and diagnosis.