Envision Summit 2025: Deep learning models of neovascular AMD

Author(s):

Majda Hadziahmetovic, MD,Geeta Lalwani, MD

At the Envision Summit 2025 in San Juan, Puerto Rico, retina program chairs Geeta Lalwani, MD, and Majda Hadziahmetovic, MD, discussed Hadziahmetovic's presentation on deep learning models of neovascular AMD as well as their time at the conference.

Video Transcript:

Editor's note: The below transcript has been lightly edited for clarity.

Geeta Lalwani, MD:

So my name is Geeta Lalwani. I'm in practice in Boulder, Colorado.

Majda Hadziahmetovic, MD:

My name is Majda Hadziahmetovic. I'm an associate professor of ophthalmology and electrical and computer engineering at Duke University.

Lalwani:

And we are the retina program chairs that put together the retina section at the Envision Summit this year, along with doctors Maria and Nina Berrocal. So Majda, this year, at Envision Summit in Puerto Rico, you spoke about the deep learning models of neovascular AMD. Can you tell us a little bit about your talk?

Hadziahmetovic:

Absolutely, first, it was a pleasure being here today and Envision seems to be getting better every year–2025 is not an exception. So what I was talking about is a deep learning framework for identification of the time point when we have to treat patients who convert to neovascular age-related macular degeneration. We know how important it is that we start the treatment in a timely manner within those two weeks, crucial two weeks, so that we can preserve good visual outcome. So this model tries to predict the time when to start the injections, and also try to predict progression of the other retinal changes over a period of time. The model is quite accurate. We still have ways to go, but very, very good initial performance has been proven already.

Lalwani:

So I asked you this during the talk, but how far in the future do you think it's going to be before we can understand how to administer treatment, to make a more personalized regimen based on this model that you talk about.

Hadziahmetovic:

That's a great question. And I think that actually we have to start thinking about what's in the black box. And I think actually neurosymbolics is going to be the way to go. That we actually start to understand a little bit better how the machines think, and what the machines taking consideration when they make this decision, to predict this is progression. Obviously, we have to have good ground truth. We have to have good data sets that actually can compare what the machine says versus what really happens in the real world and in real time.

So Geeta, I want to ask you, what do you think about this year's Envision, and what was your favorite part of it?

Lalwani:

Oh, wow. Well, easily retina was my favorite part about it. I'm very proud of what Envision Summit is, and actually, I take that back. What I've been very pleased about is the transition this year has been seamless, and the talks have been fantastic, but the kids room! I mean, that is what makes Envision Summit so unique. And I love seeing my daughter, not only in the kids room, but she sneaks into the back of the room and listens to talks, as do many other of the kids here. And it is, it is really something wonderful to be able to hear your kids laughing and screaming right across the hall, and to be able to peek at them, and, you know, be a part of their world,

Hadziahmetovic:

Indeed. And being present in the in the presentation room. I mean, kids were everywhere. And it's amazing, catching up with friends, colleagues, having this amazing scientific discussions and enjoying this beautiful weather and sun.