ATLANTA) – New findings published by Emory Eye Center neuro-ophthalmologists, Nancy J. Newman, MD and Valérie Biousse, MD, along with an international consortium of researchers from the Brain and Optic Nerve Study with Artificial Intelligence (BONSAI) group and Singapore National Eye Centre show that an artificial intelligence deep-learning system can accurately detect papilledema and other non-papilledema optic disc abnormalities from ocular fundus photographs.
Their research, “Artificial intelligence to detect papilledema from ocular fundus photographs” was published in the New England Journal of Medicine on April 14, 2020. The work was a collaborative effort between Emory Eye Center neuro-ophthalmologists and a group of researchers representing 24 centers in 15 countries around the world, led by Singapore Professor Dan Milea, MD, PhD, and Singapore engineers under the direction of Professor Tien Yin Wong, MD, PhD.
The study examines the use of a deep-learning system, or a special computer algorithm, to detect the optic disc, the visible portion of the optic nerve, and classify it as normal, papilledema (swelling of the optic disc specifically due to increased pressure in and around the brain) or another optic disc abnormality, using photographs of the back of the eye (ocular fundus).
“Examining the ocular fundus is an integral part of the physical examination that should be performed in many clinical settings where expert eye-care specialists are frequently not immediately available,” says Newman. “Recognizing abnormalities of the optic nerve is particularly important in emergency departments and neurologic and primary care clinics, where detection of papilledema can reveal vision- and life-threatening conditions of elevated intracranial pressure such as brain tumors and clots in the veins of the brain.”
“Ocular fundus photographs can remove the need for direct examination of the eye using an ophthalmoscope—an instrument non-ophthalmologists find difficult to use and rarely employ,” Newman explains. “However, just taking these photographs is not enough. Currently, someone must interpret the optic nerve appearance either on-site or via telemedicine, potentially delaying correct diagnosis and management. This artificial intelligence deep-learning system automatically and immediately correctly classifies the appearance of the optic disc without any additional clinical information.”
The research was performed using 15,846 photographs from individuals of multiple ethnicities. The study shows that the deep-learning system can accurately differentiate between abnormal optic discs and normal optic discs 99% of the time, and between papilledema and normal optic discs 98% of the time. Further studies will investigate the use of this system in a web-based application that can be used for immediate interpretation of photographs obtained in real-life settings.
“With more testing, our hope is that the Singapore engineers will eventually create a simple screening tool that is low-cost, easy to use and only requires photographs of a person’s eyes in a clinic or emergency department,” says Biousse. “This screening could allow serious neurologic problems to be quickly identified and treated, and potentially save a patient’s vision or life. Such a tool would help diagnose patients without having them travel to an eye clinic – a particularly important concern now, considering the COVID-19 pandemic.”
Newman currently serves as director of the Neuro-Ophthalmology service at the Emory Eye Center, where she maintains the LeoDelle Jolley Chair of Ophthalmology. She also holds the positions of professor of Ophthalmology and Neurology and instructor in Neurological Surgery at the Emory University School of Medicine.
Biousse is a professor in the Neuro-Ophthalmology service at the Emory Eye Center. She is the Cyrus H. Stoner professor of Ophthalmology, and she also holds the position of professor in the department of Neurology. She serves as the department of Ophthalmology’s vice-chair for Faculty Development, Diversity, Equity and Inclusion.
The research was funded by the Singapore National Medical Research Council and the SingHealth Duke-NUS Ophthalmology and Visual Sciences Academic Clinical Program.