The structural information of blood vessels in ultra-wide-angle fundus images has important guiding significance for diagnosing ophthalmic diseases. Besides, efficient and correct segmentation of ultra-wide-angle fundus vascular images has become an urgent clinical need. However, manual diagnosis is a time-consuming and laborious work. Because of the limitation of primary medical resources, the situation of missed diagnosis and misdiagnosis would probably appear in n the diagnosis of artificial experience. In this paper, we proposed an automatic vascular segmentation algorithm based on attention gate based atrous residual (AAR) Unet combined with Top-Hat transform vascular enhancement. In AAR-Unet, we integrated atrous convolution into Res-Unet to expand the receptive field and improve the correlation between objects. To improve the segmentation effect of small blood vessels under low contrast, the attention gate module is introduced. The experimental results show the overall accuracy, sensitivity, specificity and DSC of the proposed algorithm are 0.988, 0.890, 0.993 and 0.875 respectively. The results indicate the algorithm can segment the blood vessels of ultra-wide-angle fundus images accurately.

We propose a novel retinal layer segmentation method to accurately segment 10 retinal layers in optical coherence tomography (OCT) images with intraretinal fluid. The method used a fan filter to enhance the linear information pertaining to retinal boundaries in an OCT image by reducing the effect of vessel shadows and fluid regions. A random forest classifier was employed to predict the location of the boundaries. Two novel methods of boundary redirection (SR) and similarity correction (SC) were combined to carry out boundary tracking and thereby accurately locate retinal layer boundaries. Experiments were performed on healthy controls and subjects with diabetic macular edema (DME). The proposed method required an average of 415[Formula: see text]s for healthy controls and of 482[Formula: see text]s for subjects with DME and achieved high accuracy for both groups of subjects. The proposed method requires a shorter running time than previous methods and also provides high accuracy. Thus, the proposed method may be a better choice for small training datasets.