Variational Depth Estimation On Hypersphere For Panorama

Optical Coherence Tomography Angiography (OCTA) is a rapid, non-invasive imaging technique, which can display the vascular system in detail. Retinal vascular segmentation on OCTA images is of great significance for the diagnosis and treatment of many vision-related diseases. However, there is still much room for improvement in the research of retinal vascular segmentation due to the low visibility of vascular edges and high vascular complexity. Therefore, we propose a novel OCTA vascular segmentation network (VCT-Net). The network is a U-shaped network consisting of a transformer branch and a convolution branch. The structure enables the network to make full use of global and local information. The transformer branch uses a swin transformer to reduce computational complexity. Experimental results show that VCT-Net achieves better vascular segmentation performance than other deep learning methods on OCTA-6M dataset.