From the Singapore National Eye Center (D.M., D.T., S.S., C.-Y.C., T.Y.W.), Singapore Eye Research Institute (D.M., R.P.N., D.T., C.V., S.S., C.-Y.C., T.Y.W.), Duke-NUS Medical School (D.M., R.P.N., D.T., S.S., C.-Y.C., T.Y.W.), Institute of High Performance Computing, Agency for Science, Technology, and Research (J.Z., X.X., Y.L.), and Yong Loo Lin School of Medicine, National University of Singapore (S.S., T.Y.W.) - all in Singapore; Farabi Eye Hospital, Tehran University of Medical Science, Tehran, Iran (M.A.F.); the Department of Ophthalmology, Centro Hospitalar e Universitário de Coimbra, and the Coimbra Institute for Biomedical Imaging and Translational Research, University of Coimbra, Coimbra, Portugal (P.F.); the Department of Ophthalmology, Ramathibodi Hospital, Mahidol University, Bangkok, Thailand (K.V.); the Eye Center, Medical Center, University of Freiburg, Freiburg (W.A.L.), and the Department of Ophthalmology, Ruprecht Karl University of Heidelberg, Mannheim (J.B.J.) - both in Germany; IRCCS Istituto delle Scienze Neurologiche di Bologna, Unità Operativa Complessa Clinica Neurologica, and Dipartimento di Scienze Biomediche e Neuromotorie, Università di Bologna, Bologna, Italy (C.L.M.); the Department of Ophthalmology and Visual Sciences, Chinese University of Hong Kong, Hong Kong (C.Y.C.), and Zhongshan Ophthalmic Center, Sun Yat-sen University, Guangzhou (H.Y.) - both in China; the Department of Ophthalmology, Rigshospitalet, University of Copenhagen, Glostrup, Denmark (S.H.); the Department of Ophthalmology, University Hospital of Grenoble-Alpes, and Grenoble-Alpes University, HP2 Laboratory, INSERM Unité 1042, Grenoble (C.C.), Service d'Ophtalmologie, Unité Rétine-Uvéites-Neuro-Ophtalmologie, Hôpital Pellegrin, Centre Hospitalier Universitaire de Bordeaux, Bordeaux (M.-B.R.), the Department of Ophthalmology, Lille Catholic Hospital, Lille Catholic University, and INSERM Unité 1171, Lille (T.T.H.C.), the Department of Ophthalmology, University Hospital Angers, Angers (P.G.), and Rothschild Foundation Hospital, Paris (C.C.-V.) - all in France; the Department of Clinical Neurosciences, Geneva University Hospital, Geneva (N.S.); the Department of Neurology, SUNY Upstate Medical University, Syracuse, NY (L.J.M.); the American Eye Center, Mandaluyong City, Philippines (R.K.); Moorfields Eye Hospital NHS Foundation Trust and UCL Institute of Ophthalmology, University College London, London (P.Y.-W.-M.), and Cambridge Eye Unit, Addenbrooke's Hospital, Cambridge University Hospitals, and Cambridge Centre for Brain Repair and Medical Research Council Mitochondrial Biology Unit, Department of Clinical Neurosciences, University of Cambridge, Cambridge (P.Y.-W.-M.) - all in the United Kingdom; the Save Sight Institute, Faculty of Health and Medicine, University of Sydney, Sydney (C.L.F.); the Department of Ophthalmology and Neurology, Mayo Clinic, Rochester, MN (J.J.C.); the Department of Neuro-ophthalmology, Sankara Nethralaya, Medical Research Foundation, Chennai, India (S.A.); the Departments of Ophthalmology, Neurology, and Neurosurgery, Johns Hopkins University School of Medicine, Baltimore (N.R.M.); and the Departments of Ophthalmology and Neurology, Emory University School of Medicine, Atlanta (N.J.N., V.B.).
N Engl J Med. 2020 Apr 30;382(18):1687-1695. doi: 10.1056/NEJMoa1917130. Epub 2020 Apr 14.
Nonophthalmologist physicians do not confidently perform direct ophthalmoscopy. The use of artificial intelligence to detect papilledema and other optic-disk abnormalities from fundus photographs has not been well studied.
We trained, validated, and externally tested a deep-learning system to classify optic disks as being normal or having papilledema or other abnormalities from 15,846 retrospectively collected ocular fundus photographs that had been obtained with pharmacologic pupillary dilation and various digital cameras in persons from multiple ethnic populations. Of these photographs, 14,341 from 19 sites in 11 countries were used for training and validation, and 1505 photographs from 5 other sites were used for external testing. Performance at classifying the optic-disk appearance was evaluated by calculating the area under the receiver-operating-characteristic curve (AUC), sensitivity, and specificity, as compared with a reference standard of clinical diagnoses by neuro-ophthalmologists.
The training and validation data sets from 6779 patients included 14,341 photographs: 9156 of normal disks, 2148 of disks with papilledema, and 3037 of disks with other abnormalities. The percentage classified as being normal ranged across sites from 9.8 to 100%; the percentage classified as having papilledema ranged across sites from zero to 59.5%. In the validation set, the system discriminated disks with papilledema from normal disks and disks with nonpapilledema abnormalities with an AUC of 0.99 (95% confidence interval [CI], 0.98 to 0.99) and normal from abnormal disks with an AUC of 0.99 (95% CI, 0.99 to 0.99). In the external-testing data set of 1505 photographs, the system had an AUC for the detection of papilledema of 0.96 (95% CI, 0.95 to 0.97), a sensitivity of 96.4% (95% CI, 93.9 to 98.3), and a specificity of 84.7% (95% CI, 82.3 to 87.1).
A deep-learning system using fundus photographs with pharmacologically dilated pupils differentiated among optic disks with papilledema, normal disks, and disks with nonpapilledema abnormalities. (Funded by the Singapore National Medical Research Council and the SingHealth Duke-NUS Ophthalmology and Visual Sciences Academic Clinical Program.).
非眼科医生不能自信地进行直接检眼镜检查。利用人工智能从眼底照片中检测视盘水肿和其他视盘异常的方法尚未得到充分研究。
我们训练、验证和外部测试了一个深度学习系统，该系统可从 15846 张回顾性采集的眼部照片中分类视盘是否正常、是否有视盘水肿或其他异常，这些照片是通过药物性瞳孔扩张和多种数字相机在来自多个种族人群的患者中获得的。在这些照片中，来自 11 个国家 19 个地点的 14341 张用于训练和验证，来自 5 个其他地点的 1505 张用于外部测试。通过计算受试者工作特征曲线（ROC）下的面积（AUC）、敏感性和特异性来评估该系统对视盘外观进行分类的性能，并与神经眼科医生的临床诊断参考标准进行比较。
来自 6779 名患者的训练和验证数据集包括 14341 张照片：9156 张正常视盘，2148 张视盘水肿，3037 张其他异常视盘。各站点分类为正常的百分比范围为 9.8%至 100%；各站点分类为视盘水肿的百分比范围为 0 至 59.5%。在验证集中，该系统以 AUC 为 0.99（95%CI，0.98 至 0.99）区分视盘水肿与正常视盘，以 AUC 为 0.99（95%CI，0.99 至 0.99）区分正常与非视盘水肿异常。在外部测试数据集的 1505 张照片中，该系统检测视盘水肿的 AUC 为 0.96（95%CI，0.95 至 0.97），敏感性为 96.4%（95%CI，93.9 至 98.3），特异性为 84.7%（95%CI，82.3 至 87.1）。
使用药物性瞳孔扩张眼底照片的深度学习系统可区分视盘水肿、正常视盘和非视盘水肿异常。（由新加坡国家医学研究委员会和新加坡保健集团杜克-国大学眼科和视觉科学学术临床项目资助）。
