【摘要】 为解决目前临床上对早期糖尿病视网膜病(diabetic retinopathy,DR)诊断困难的问题,利用多焦视网膜电流图(multifocal electroretinogram,mfERG)和模式分类的方法,对DR进行早期诊断和自动分类的研究。对31例正常样本和58例早期DR样本,提取mfERG信号的平均一阶响应(K1成分)峰值和模板相关系数两个关键特征。使用线性分类器进行分类,并用“留一法”统计结果,正常人和早期DR病例的分类错误率为21.35%。该文还用聚类方法分析了早期DR样本的分布特点。结果显示,使用平均一阶响应的N2峰的幅值和位置作为特征对早期DR的诊断效果较好;早期DR样本的聚类结果具有一定的临床价值。更多还原

【Abstract】 Pattern classification was used with multifocal electroretinogram (ERG) samples for early detection of diabetic retinopathy (DR). Two key features, the amplitude and time of peaks for the average first-order kernel and the correlation coefficient between samples and templates, were extracted from 31 normal samples and 58 early stage DR samples. The classification error rate for normal and early stage DR samples reached 21.35% using a linear classifier and the leave-one-out method. The cluster method was also used to calculate the distribution of the early DR samples. The results show that the N2 peak of the first-order kernel provides the best early detection of DR, with the cluster results having some clinical value.更多还原