The following transcript has been edited for clarity.

Hi, my name is Cecilia S. Lee, MD, MS. I'm the Jane Hardesty Poole Distinguished Professor of Ophthalmology and Visual Sciences at Washington University in St. Louis.

And I'm Aaron Y. Lee, MD, MSCI. I'm the new department chair for the Department of Ophthalmology at Washington University in St. Louis.

Retinal Physician: What are the biggest barriers to AI in ophthalmology?

Dr. Aaron Lee: Some of the biggest barriers that we've had to deal with in our journey initially had to do with the standardization of a lot of the imaging devices that were present in our field. So unlike in radiology where there’s these 2 major players that just had to agree on how to store their images, in our field, we have both a blessing and a curse in that there's so much innovation going on. There's so many different companies with many different devices, but the curse side of that is that we don't have good ways to standardize that data and store it in a way that we can pull them down efficiently at scale to train and build these complex AI models. So, for many, many years in our journey, we struggled with this, and we had to work a lot with the manufacturers.

Now we're moving into this era where the manufacturers are indeed implementing things like the Digital Imaging and Communications in Medicine (DICOM) standard and exchange standards like Fast Healthcare Interoperability Resources (FHIR) from electronic health records (EHR) are being talked about more and more. And so we're finally moving the field of ophthalmology into this era where we can start to see the collation of large data sets to drive new insights using AI become a lot easier.

Dr. Cecilia Lee: I think another aspect is that obviously, once you have that data, to maximize the benefit or the strength of the data is really to be able to share. And then data sharing while really protecting the privacy of the participants’ data has been a huge challenge. And so we came up with a Swiss cheese mechanism in terms of protecting the privacy of the participants and creating a license that specifically prohibits any efforts or attempts to reidentify a participant or use the data in a harmful way for any vulnerable population.

I think additionally, right now, we're really talking a lot about trust in the AI and the AI models and the data. So I think getting community engagement on board, so community partners and also many different stakeholders in developing the data set and sharing the data set would be really helpful. So that has been a major component of our work. 

Dr. Aaron Lee: And just to amplify what Cecilia said, we were sort of witnessing these waves of AI papers coming out, and we realized there was this critical gap of really high-quality data sets that people could use to train AI models. And so that's where we developed this project called AI-READI. And we shared this data set for the world. I think we’ve shared it with more than a thousand groups at this point, where people can download the data set to their computers to do analytics and train deep-learning models with.

So, we're trying to fill the gaps that we see in the field. The next gap that we sort of are tackling is this idea of, how do we get these AI models into the clinic? How do we get them deployed safely so that we can really start to transform the way that ophthalmology care is delivered? 

Dr. Cecilia Lee: We mentioned publicly accessible data sets, so that people are able to download the data set after doing the presentation and doing some work. But there is another set that is controlled access, where you require a Data Use Agreement (DUA) and a more traditional mechanism. I think because the publicly accessible data set is downloadable and is really easily accessible, we have been able to make all these moves with many partners already publishing research. RP