摘要
在青光眼诊断中,眼底照和光学相干断层扫描(OCT)是最主要的两种眼科检查手段。对于眼底照和OCT图像数据,首先设计基于专家知识的机器学习算法,提取杯盘比曲线和视神经纤维层厚度曲线的尺度和形态特征,进而提出一种基于Dempster-Shafer(DS)证据推论的多视图集成学习方法,利用支持向量机(SVM)和逻辑回归进行青光眼预测。在一个真实数据集合上,对所提方法的预测性能进行了评估实验,结果表明,本文算法与眼科专家出具的诊断结果高度一致,并且比已有算法有更好的敏感性、特异性和更高的预测准确率。
In the diagnosis of glaucoma, fundus images and optical coherence tomography(OCT) are the two types of eye examination that doctors pay most attention to. To enhance the analysis of fundus images and OCT reports which are also images, we have designed a machine learning algorithm based on expert knowledge to extract the scale and morphological characteristics of cup-to-disk ratio curves and retinal nerve fiber layer thickness curves. Furthermore, we propose a multi-view integrated learning method based on Dempster-Shafer(DS) evidence theory, and make glaucoma predictions based on support vector machines(SVM) and logistic regression. Finally, we evaluated the predictive performance of the proposed method using a real data set. The experimental results show that the proposed algorithm is highly consistent with the diagnostic results given by ophthalmologists, and achieves higher sensitivity, specificity and accuracy than existing algorithms.
引文
[1] KLEIN B E K,KLEIN R,SPONSEL W E,et al.Prevalence of glaucoma—the beaver dam eye study[J].Ophthalmology,1992,99(10):1499- 1504.
[2] SALAM A A,AKRAM M U,WAZIR K,et al.Autonomous glaucoma detection from fundus image using cup to disc ratio and hybrid features[C]//2015 IEEE International Symposium on Signal Processing and Information Technology (ISSPIT).Abu Dhabi,2015:370- 374.
[3] DAS P,NIRMALA S R,MEDHI J P.Diagnosis of glaucoma using CDR and NRR area in retina images[J].Network Modeling Analysis in Health Informatics & Bioinformatics,2016,5(1):1- 14.
[4] XU Y L,JIA X,HU M,et al.Feature extraction from optic disc and cup boundary lines in fundus images based on ISNT rule for glaucoma diagnosis[J].Journal of Medical Imaging & Health Informatics,2015,5(8):1833- 1838.
[5] BIZIOS D,HEIJL A,HOUGAARD J L,et al.Machine learning classifiers for glaucoma diagnosis based on classification of retinal nerve fibre layer thickness parameters measured by stratus OCT[J].Acta Ophthalmologica,2010,88(1):44- 52.
[6] GROUP A I F G S.Combining information from three anatomic regions in the diagnosis of glaucoma with time-domain optical coherence tomography[J].Journal of Glaucoma,2008,23(3):129- 135.
[7] LARROSA J M,POLO V,FERRERAS A,et al.Neural network analysis of different segmentation strategies of nerve fiber layer assessment for glaucoma diagnosis[J].Journal of Glaucoma,2015,24(9):672- 678.
[8] SILVA F R,VIDOTTI V G,CREMASCO F,et al.Sensitivity and specificity of machine learning classifiers for glaucoma diagnosis using spectral domain OCT and standard automated perimetry[J].Arquivos Brasileiros De Oftalmologia,2013,76(3):170- 174.
[9] GOPINATH K,SIVASWAMY J,MANSOORI T.Automatic glaucoma assessment from angio-OCT images[C]//IEEE 13th International Symposium on Biomedical Imaging (ISBI).Prague,2016:193- 196.
[10] ZADEH L A .A simple view of the dempster-shafer theory of evidence and its implication for the rule of combination[J].Ai Magazine,1986,7(2):85- 90.
[11] ZHANG C,MA Y.Ensemble machine learning methods and applications[M].New York:Springer Science & Business Media,2012:1- 34.
[12] CORTES C,VAPNIK V.Support-vector networks[J].Machine Learning,1995,20(3):273- 297.
[13] COX D R.The regression analysis of binary sequences[J].Journal of the Royal Statistical Society,1958,20(2):215- 242.
[14] PLATT J C.Probabilities for SV machines[M]//SMOLA A J,BARTLETT P,SCHOLKOPF B.Advances in large margin classifiers.Cambridge:The MIT Press,1999:61- 74.