Vascular segmentation and reconstruction in diabetic retinopathy based on deep learning
A method capable of retinal vessel segmentation and three-dimensional(3D)reconstruction is proposed for the early diagnosis of diabetic retinopathy.The 3D reconstruction can avoid the misjudgments of blood vessel length,curvature and branch angle after segmentation,which will affect the early diagnosis.IAAnet algorithm for retinal image segmentation combines traditional Unet with Inception V3,atrous spatial pyramid pooling and AttentionGates to reduce information loss and avoid over-fitting,thereby improving the network's ability to extract features.The projection reconstruction method is used to restore the 3D information of blood vessels,and supports the adjustments of brightness and contrast,so that doctors can better observe the real state of blood vessels.The proposed algorithm has an accuracy,recall rate,F1 score,intersection over union and area under ROC curve of 97.68%,96.07%,97.26%,92.79%and 94.00%,respectively.Compared with other networks,IAAnet algorithm exhibits higher segmentation accuracy,and can obtain more vascular information in 3D image after 3D projection reconstruction to assist in the early diagnosis.
deep learningdiabetic retinopathyInception V3attention gateatrous spatial pyramid pooling3D projection reconstruction
万方数据
维普
