GS-Net:U-Net with Gated Spatial Convolution for Retinal Image Segmentation

Optical Coherence Tomography Angiography (OCTA) plays a crucial role in the early diagnosis and continuous monitoring of eye diseasesThis study presents GS-Net, aimed at accurately segmenting various retinal structures, including capillaries, arteries, veins, and the foveal avascular zone (FAZ). The core of GS-Net is an improved U-Net based on gated spatial convolution. The advantages of gated spatial convolution include dynamic feature selection and enhanced representation capabilities. Extensive experiments demonstrate that GS-Net outperforms existing state-of-the-art methods on two subsets of the publicly available OCTA-500 dataset. Compared to Intersection over Union (IoU), the average IoU scores are 77.91% and 83.68%, respectively. The proposed GS-Net shows excellent performance in OCTA image segmentation, providing an efficient and accurate multi-task segmentation solution for ophthalmic diagnosis.