【摘要】 糖尿病性黄斑水肿(diabetic macular edema, DME)是视网膜上毛细血管渗漏的液体在细胞间隙聚集引起的中央视网膜肿胀。基于眼底光学相干断层扫描(optical coherence tomography,OCT)图像的黄斑区积液检测在DME治疗方面具有重要的作用。因此，提出了一种分割网络SR-Net，以实现在OCT图像上自动分割视网膜内积液(intraretinal fluid,IRF)。SR-Net以Res2Net为主干，并引入了空间通道压缩和激发(spatial and channel squeeze & excitation, scSE)模块、平行解码器等结构。基于Kermany数据集的实验结果表明，SR-Net分割IRF的平均Dice系数、灵敏度、特异性分别为75.20%、77.56%、99.98%，最大Dice系数可达88.19%，证明了SR-Net的有效性。更多还原

【Abstract】 Diabetic macular edema(DME) is a swelling of the central retina caused by fluid leaking from capillaries that collects in the intercellular space. The detection of macular effusion based on optical coherence tomography(OCT) images of fundus plays an important role in DME treatment. Therefore, a segmentation network SR-Net is proposed to automatically segment intraretinal fluid(IRF) on OCT images. SR-Net is based on Res2Net, and introduces spatial and channel squeeze & excitation(scSE) module, parallel decoder and other structures. The experimental results based on Kermany dataset shows that the average Dice coefficient,sensitivity and specificity of IRF segmentation based on SR-Net are 75.20%, 77.56% and 99.98%, respectively,and the maximum Dice coefficient is up to 88.19%, which proves the effectiveness of SR-Net.更多还原