RELATED CONTENT

Some eyecare practitioners are incorporating AI technology into their practices to streamline rote administrative tasks so they can spend more time with patients. The clinical application of AI technology in ophthalmology has not yet been widely adopted.



Alvin Liu, MD



“If you look at the research in AI and ophthalmology there is a lot of progress,” said Alvin Liu, MD, an assistant professor of ophthalmology and the founding director of the Wilmer Precision Ophthalmology Center of Excellence at Johns Hopkins Medicine. “And if you look at the FDA-approved products, there are very few. And if you look at the actual usage of it, it’s in its infancy.”

According to Dr. Liu, one of the most common uses of AI in a clinical setting is the detection of diabetic retinopathy in retinal photographs, typically done in a primary care physician’s office. If abnormalities are found, they are sent to an ophthalmologist for review. Additionally, AI may be used in optical coherence tomography (OCT), to screen for age-related macular degeneration (AMD) or diabetic retinopathy.

“People use the word AI really loosely nowadays,” said Dr. Liu. “Generative AI, the technology related to Chat GPT, which is the real AI, has been deployed in the revenue cycle/management side of things when it comes to preauthorization and coding. I am involved in those efforts as well and they don’t require FDA approval. It’s not widespread in ophthalmology but is definitely being used in other fields of medicine already.”





Given that ophthalmology is generally outpatient-based, and issues are fixed “on the spot in the clinic,” Dr. Liu said there may be less need for AI tools like generative AI to craft patient communication. But ultimately, AI technology is a work in progress in a clinical setting.

“There are two major obstacles to the scaling of AI when it comes to clinical decision-making,” said Dr. Liu. “Number one is the regulatory hurdle is quite high for the FDA to get that approval, much higher than CE marking in Europe, so that is going to impede the adoption.

“The second thing is, a lot of these AI tools, even for the clinical decision support ones, do work well scientifically, but often times when you deploy them in the real world, we see a drop in performance, or a difference in performance from the original studies. These AI tools are just tools. They need to be reworked into the current workflow. A lot of work needs to be done on the implementation side. Without the right workflow and change management in place, we won’t be able to use them and if we can’t demonstrate the ROI of these tools, the decision-makers won’t deploy them.”

- Stefani Kim, Senior Editor, Lenses & Technology