摘要
硬渗出液是早期糖尿病性视网膜病变(diabetic retinopathy,DR)的主要病症之一,在眼底图像中占据的像素点较少,其检测容易受视盘、软渗出液的干扰。针对这些问题,在U型网络(U-Net)结构的基础上,通过在编码器和解码器中融入残差模块和残差通道注意力模块学习硬渗出液的细微特征,在跳跃连接中加入一种新的多尺度通道注意力(multi-scale channel attention,MSCA)模块提升网络对稀疏小病灶的分割能力,提出了MSCA U-Net。基于超广角眼底图像数据集和印度糖尿病性视网膜病变图像数据集的实验结果表明,与其他基于卷积神经网络的图像分割方法相比,所提方法具有更高的硬渗出液分割精度。
Hard exudates are one of the main diseases of early diabetic retinopathy(DR).Hard exudates occupy less pixels in the fundus image,and the detection of hard exudates is easily disturbed by the optic disc and soft exudate.To solve these problems,based on the U-Net structure,the residual module and residual channel attention module are integrated into the encoder and decoder to learn the subtle features of hard exudates,a new multi-scale channel attention(MSCA)module is added to the skip connection to improve the ability of the network to segment sparse small lesions,and the MSCA U-Net is proposed.Experimental results based on ultra-wide-angle fundus image dataset and India diabetic retinopathy image dataset show that the proposed method has higher segmentation accuracy of hard exudates than other image segmentation methods based on convolutional neural networks.
作者
傅迎华
张葛
左嵩
FU Yinghua;ZHANG Ge;ZUO Song(School of Optical-electrical and Computer Engineering,University of Shanghai for Science and Technology,Shanghai 200093,China;Information Management Department,Xinhua Hospital Affiliated to Shanghai Jiaotong University School of Medicine,Shanghai 200000,China)
出处
《控制工程》
CSCD
北大核心
2024年第7期1244-1253,共10页
Control Engineering of China
基金
教育部产学合作协同育人项目(202102026003,202101315001)。
