DPANet: Automated Segmentation of Optic Disc with Optical Coherence Tomography Angiography in Ophthalmology : Subtitle: Segmentation of Optic Disc with OCTA

Automatic medical image segmentation plays an important role in medical diagnosis. Although machine-learning methods have achieved superior performance in many image segmentation tasks, there exist still some limitations, including the accurate segmentation of edge regions and lack of data sets. In clinical practice, accurate segmentation of the optic disc is important for disease diagnosis. We aim to design a novel convolutional network for optic disc segmentation in OCTA fundus images. To address these problems, we propose a novel framework, namely Dual-Path Attention Network (DPANet) for OCTA fundus image segmentation. We constructed a new Optic Disc segmentation OCTA Image Dataset (ODOID). ODOID consists of 50 OCTA images with optic disc annotations by pixel level. Experimental results show that our framework is effective and achieves the best results on the new ODOID dataset. The new dataset and the code are available at https://17861318579.github.io/ODOID.