摘要
青光眼性视盘改变、视网膜出血和渗出是诊断眼底疾病的主要特征,传统方法可诊断是否患有眼底疾病,但难以对眼底疾病诊断结果给出合理的解释。鉴于此,提出了一种融合双重注意力机制的卷积神经网络(CNN),实现了眼底多病变特征的自动诊断。CNN采用残差结构,在残差块中利用分组卷积以减少网络参数量,并在每组卷积之后嵌入通道和空间注意力机制以提升眼底病变诊断的准确率。该模型在宁波市眼科医院临床数据上进行了实验,青光眼性视盘改变、视网膜渗出和出血3种病变的诊断准确率分别为98.17%、97.49%、97.15%,结果表明:该模型在眼底多病变诊断中表现出很好的特征提取能力和诊断性能。
Glaucomatous optic disc changes,retinal hemorrhage and exudates are the main characteristics of fundus disease diagnosis.Traditional methods can diagnose fundus disease,but it is difficult to give a reasonable explanation for the diagnoses of fundus disease.Therefore,a convolutional neural network(CNN)fuses dual attention mechanism is proposed to realize the automatic diagnosis of multiple fundus lesions.CNN adopts residual structure and uses grouping convolution in residual block to reduce the number of network parameters.After each group of convolution,channel and spatial attention mechanisms are embedded to improve the accuracy of fundus lesion diagnosis.The model is tested on the clinical data of Ningbo Eye Hospital,and the diagnostic accuracy of glaucomatous optic disc changes,retinal exudation and hemorrhage are 98.17%,97.49%and 97.15%,respectively.The results show that the model shows good feature extraction ability and diagnostic performance in the diagnosis of multiple fundus lesions.
作者
蒋杰伟
郭刘飞
巩稼民
强薇
吴成超
李中文
JIANG Jiewei;GUO Liufei;GONG Jiamin;QIANG Wei;WU Chengchao;LI Zhongwen(School of Electronic Engineering,Xi’an University of Posts and Telecommunications,Xi’an 710121,China;School of Communication and Information Engineering,Xi’an University of Posts and Telecommunications,Xi’an 710121,China;Ningbo Eye Hospital,Wenzhou Medical University,Ningbo 315000,China)
出处
《传感器与微系统》
CSCD
北大核心
2023年第5期152-155,160,共5页
Transducer and Microsystem Technologies
基金
国家自然科学基金资助项目(61775180)
国家重点研发计划资助项目(2018YFC0116500)
宁波市科技计划资助项目(2019C50045)
高校青年教师科研基金资助项目(205020022)。
关键词
通道注意力机制
分组卷积
空间注意力机制
多病变诊断
channeled attention mechanism
grouping convolution
spatial attention mechanism
multi-lesion diagnosis