Related Experiment Video
Updated: Jun 20, 2026
Using Retinal Imaging to Study Dementia
Published on: November 6, 2017
Deep Learning-Based Detection of Papilledema on Retinal Photographs From Handheld Cameras: A Prospective Study
Ayse Gungor1, Zhiqun Tang, Jing L Loo
1Sorbonne Université (AG, IS), Institut du Cerveau-Paris Brain Institute-ICM, Inria, Inserm, CNRS, APHP, Hôpital de la Pitié Salpêtrière, Paris, France; Rothschild Foundation Hospital (AG, LT, IS, DM), Neuro-Ophthalmology Department, Rothschild Computational and Visual Neurosciences Laboratory, Paris, France; Department of Ophthalmology (ZT, JLL, SS), Yong Loo Lin School of Medicine, National University of Singapore, Singapore; Singapore National Eye Centre (JLL, STLC, SS, RFCM, DM), Singapore; Duke-NUS Medical School (JLL, STLC, SS, DM), Singapore; Visual Neurosciences Group (SS, RPN, DM), Singapore Eye Research Institute, Singapore; Departments of Ophthalmology (NJN, VB) and Neurology (NJN, VB), Emory University School of Medicine, Atlanta, Georgia; and Copenhagen University (DM), Copenhagen, Denmark.
A deep learning system (DLS) accurately identifies papilledema and other optic neuropathies from retinal images. This AI tool shows high performance in real-world clinical settings for neuro-ophthalmology diagnosis.
Area of Science:
- Ophthalmology
- Artificial Intelligence
- Medical Imaging
Background:
- Papilledema and optic neuropathies are critical neuro-ophthalmologic findings requiring prompt diagnosis.
- Retinal photograph analysis is essential for identifying these conditions.
Purpose of the Study:
- To evaluate a deep learning system's (DLS) performance in identifying papilledema and other optic neuropathies.
- To assess the DLS on prospectively acquired, nonmydriatic retinal images from a neuro-ophthalmology department.
Main Methods:
- A multi-center study used 20,533 retinal photographs from 10,647 patients.
- A DLS was trained on mydriatic images and tested on nonmydriatic images from a handheld camera.
- The DLS classified images as papilledema, other optic disc abnormalities, or normal, with performance measured by AUC, sensitivity, specificity, and accuracy.
Main Results:
- The DLS achieved 99.5% accuracy, 81.0% sensitivity, 99.7% specificity, and 98.3% AUC in differentiating papilledema from other conditions and normal controls.
- Performance was evaluated at both eye and patient levels.
Conclusions:
- A DLS trained on mydriatic photographs demonstrates excellent diagnostic performance for optic disc abnormalities.
- The system is effective when applied to nonmydriatic retinal images captured with handheld cameras in clinical practice.
