Tiny yet efficient macrostructure-assisted U-Net for OCTA retinal vessel segmentation

Optical coherence tomography angiography(OCTA), as a noninvasive imaging technique, has the ability to demonstrate high-resolution retinal vessels. In this study, we propose a tiny yet efficient neural network for accurate retinal vessel segmentation in OCTA images, with specially focusing on the accurate micro-structure segmentation and the size of the network. The designed neural network employs a size-reduced U-Net and an additional full-resolution branch is introduced to better extract features of macro-structures in an image. Throughout the study, we focus on optimizing the segmentation effect while paying attention to the balance between the number of model parameters and inference speed. With this comprehensive design, we achieved the task of segmenting retinal blood vessels in OCTA images more accurately in less time. This work provides useful methods and ideas for further research in the field of medical image processing in the future.