AI Used to Dx Optic Nerve Abnormalities

Download PDF

An international neuro-ophthalmologists successfully harnessed artificial intelligence (AI) to detect optic nerve abnormalities from photographs taken with a variety of commercially available digital fundus cameras.¹ Their AI algorithm used deep learning neural networks to distinguish papilledema from other optic neuropathies as well as from normal optic discs.

“This system is intended to help general physicians and nonophthalmic health care providers who need an accurate and immediate assessment of the optic nerve head, in the absence of an ophthalmologist,” said Tien Y. Wong, MD, PhD, at the Singapore National Eye Centre and Duke-National Uni­versity of Singapore Medical School. He is a member of the Brain and Optic Nerve Study with Artificial Intelligence (BONSAI) consortium, which created the diagnostic system.

Training and validation. Neuro-oph­thalmologists at 19 sites in 11 countries read 14,341 digital color ocular fundus photographs collected from a multi­ethnic population (Indian, Asian, and non-Asian patients). From the fundus images, they retrospectively diagnosed 9,156 normal discs, 2,148 discs with papilledema, and 3,037 discs with other abnormalities. They then trained the system to do the same.

Next, they externally tested the system’s performance on 1,505 pho­tographs at five additional sites in five countries. The AI system correctly identified 96.4 of every 100 fundus images with papilledema and 84.7 of every 100 fundus images without papilledema.

Classification errors. The system was not always correct. Of 360 discs with papilledema, 15 (4.2%) were misclassified as “other abnormalities.” However, the system never misread the abnormal discs as normal.

Still investigational. Though the system has been validated in the five external testing cohorts, it must re­ceive regulatory approval in different countries, Dr. Wong said. Moreover, a number of issues need to be resolved, including medicolegal concerns regard­ing liability for a wrong diagnosis.

To address these and other ques­tions, the group is conducting further prospective, real-life studies in Singapore and elsewhere. “If proven efficient, this system could represent an important step in decision-making processes of ordering brain imaging and/or lumbar punctures,” said BONSAI principal investi­gator Dan Milea, MD, PhD, also at the Singapore Eye Centre and Duke-National University of Singapore Med­ical School. Moreover, he said, the use of such a system could reduce the incidence of unnecessary or expensive investigations—and spare patients any associated dis­comfort.

—Miriam Karmel

___________________________

1 Milea D et al. New Engl J Med. 2020;382(18):1687-1695.

___________________________

Relevant financial disclosures—Dr. Milea: None. Dr. Wong: Allergan: C; Bayer: C; Boehringer-Ingelheim: C; Genentech: C; Merck: C; Novartis: C; Oxurion: C; Roche: C; Samsung: C.

For full disclosures and the disclosure key, see below.

• In 10 Days: From Skin Cells to Photoreceptors

• Cataract Surgery Safe in Patients With Heart Failure

• Real-World Study of Brolucizumab Finds Severe Retinal Vasculitis