改进的U-Net在OCT视网膜分割中的应用

Improved U-Net for segmentation of OCT retinal images

视网膜眼底图像能够反映出视网膜的剖面形态,从而能够更加精确地对视网膜进行解剖。图层的划分与定量测算对于很多视网膜及视神经的疾病都是非常有用的,因此对眼底图像进行分割具有很强的现实意义。本文利用SENet模块对编码器为Vgg-16的U-Net引入注意力机制,实现了对视网膜眼底图像的精确分割。实验表明,改进的网络在性能上优于其他两个基础模型。

The retinal fundus image can reflect the section shape of the retina,so as to be able to dissect the retina more accurately,and the division and quantitative measurement of layers are for many retinal and optic nerve diseases,so the segmentation of fundus images has strong practical significance.And,this paper uses the SENet module to introduce the attention mechanism into the U-Net with encoder Vgg-16,and realizes the accurate segmentation of retinal fundus images.Experiments show that the improved network outperforms the other two basic models in performance.