American Academy of Ophthalmology Meeting

American Academy of Ophthalmology Meeting

October 14, 2019
2 min read

AREDS2 formulation, Mediterranean diet and deep learning may improve diagnosis, treatment of AMD

You've successfully added to your alerts. You will receive an email when new content is published.

Click Here to Manage Email Alerts

We were unable to process your request. Please try again later. If you continue to have this issue please contact

Emily Y. Chew

SAN FRANCISCO —The AREDS2 formulation and a Mediterranean diet, especially fish, can help reduce the risk for patient progression to advanced age-related macular degeneration, while advancements in deep learning capabilities can improve ophthalmic diagnostic capabilities, according to a speaker here.

“The AREDS2 formulation is recommended to patients with intermediate or late AMD, and genetic testing is encouraged for research purposes but not for clinical care yet. Deep learning may help us in the future for more accurate and more efficient clinical practice but won’t replace us,” Emily Y. Chew, MD, said during the Jackson Memorial Lecture at the American Academy of Ophthalmology annual meeting.

The AREDS2 formulation, consisting of vitamin C, vitamin E, lutein/zeaxanthin, zinc and copper, is recommended for patients with intermediate or late AMD in one eye to reduce progression to advanced AMD. Additionally, a high adherence to the Mediterranean diet is associated with reduced risk for progression to late AMD by 25% to 40%, Chew said.

Chew noted fish is the component of the Mediterranean diet associated with a 31% decreased risk for progression to geographic atrophy.

However, patient genetics can also increase risks for AMD progression. Chew said patients with the presence of complement factor H exhibited a sevenfold increased risk for late AMD.

“Genetic testing may determine the pathways and help us identify patients at high risk of disease progression,” she said.

Deep learning systems may also improve diagnostic capabilities of AMD severity and progression risk. Chew and colleagues developed DeepSeeNet, a deep learning model for automatic classification of patient-based AMD severity. The deep learning model assessed features of AMD on color fundus photographs and classified severity of AMD using the AREDS simplified severity scale to predict the risk for late AMD progression.

In almost all instances, the model was superior to human gradings for detecting drusen, including size, and pigmentary changes. However, the performance by the model for late AMD was not superior to human gradings.

“We think part of the problem was we didn’t have enough numbers of late AMD to train our training set. ... Finally, the overall accuracy of the whole system was by far better with DeepSeeNet. The deep learning was by far more superior than that of the retinal specialists,” she said. – by Robert Linnehan



Chew EY. Jackson Memorial Lecture: Age-related macular degeneration: Nutrition, genes and deep learning. Presented at: American Academy of Ophthalmology annual meeting; October 11-15, 2019; San Francisco.

Disclosure: Chew reports no relevant financial disclosures.