正常角膜和圆锥角膜的特征提取

发布时间：2018-06-27 23:45

本文选题：角膜 + 圆锥角膜　；参考：《光学精密工程》2015年10期

【摘要】：基于角膜测量仪器Corvis ST采集的图像视频,提出提取新特征参数以便准确区分正常角膜和圆锥角膜。首先对图像进行滤波、分割等预处理,检测角膜上下边界,并计算前角膜曲率值;用小波变换分析角膜曲率变化,获取与角膜运动趋势相关的特征,包括角膜运动的整体趋势和角膜振动的范数和标准差。然后,基于均方误差最小化法,提取特征参数,构建最优参数。最后,用支持向量机(SVM)对正常角膜和圆锥角膜进行分类。从频率的角度实施的实验显示角膜在基本运动趋势上存在着振动过程。此外,提出的参数优于形变幅度(DA)、峰值距离(PD)等传统参数,使准确度、灵敏度和特异性分别提高了10.2%,5.7%和6.9%。受试者工作特征曲线(ROC)下面积为0.948,接近于1。结果显示本文方法自动提取的特征参数可提高正常角膜和圆锥角膜区分的准确性,对临床诊断有辅助作用。
[Abstract]:Based on the image and video collected by Corvis St, a new feature parameter is proposed to distinguish the normal cornea from the keratoconus accurately. First of all, the image is pre-processed, such as filtering, segmentation and so on, the upper and lower corneal boundaries are detected, and the curvature of the anterior cornea is calculated. The wavelet transform is used to analyze the changes of corneal curvature to obtain the features related to the corneal movement trend. Including the overall trend of corneal movement and corneal vibration norm and standard deviation. Then, based on the mean square error minimization method, the feature parameters are extracted and the optimal parameters are constructed. Finally, support vector machine (SVM) is used to classify normal cornea and conus cornea. Experiments carried out from the angle of frequency show that there is a vibration process in the basic motion trend of cornea. In addition, the proposed parameters are superior to the traditional parameters such as deformation amplitude (DA) and peak distance (PD). The accuracy, sensitivity and specificity of the proposed parameters are increased by 10.2% and 6.9%, respectively. The area under the operating characteristic curve (ROC) was 0. 948, close to 1. The results show that the automatic extraction of characteristic parameters can improve the accuracy of distinguishing normal cornea from keratoconus, and it is helpful for clinical diagnosis.
【作者单位】：复旦大学电子工程系;医学影像计算及计算机协助介入重点实验室;香港理工大学跨学科生物医学工程部门;首都医科大学附属北京同仁医院北京市眼科研究所;中国人民解放军总医院眼科系;
【基金】：国家973重点基础研究发展计划资助项目(No.2015CB755500) 国家自然科学基金资助项目(No.61471125,No.81271052) 香港理工大学内地联合监督计划资助项目(No.G-UB58)
【分类号】：TP391.41;R772.2

【参考文献】