In this episode of Interventional Mindset, Nathan Radcliffe, MD, discusses ways ophthalmologists can use artificial intelligence (AI) in their day-to-day workflows. He also provides deeper insights into using AI to assist with optical coherence tomography (OCT) analysis in glaucoma.
“It's been amazing to me how many practical things related to the practice of ophthalmology that AI can be helpful with.”
Dr. Radcliffe is a board-certified ophthalmologist at New York Ophthalmology with offices in the South Bronx, Washington Heights, and Jamaica. He’s an expert in cataract and glaucoma care and specializes in medical, laser, and surgical treatments.
AI in ophthalmology practices fast facts
- Ophthalmologists can use AI to research clinical questions and access medical information faster.
- Documentation, patient education materials, and administrative tasks can be completed more efficiently with AI.
- AI tools can save ophthalmologists time and energy by assisting with daily responsibilities such as proofreading and dictation.
- AI can make it easier to evaluate complex cases and narrow differential diagnoses.
- De-identified OCT images and visual field data can be analyzed by AI to support glaucoma assessment and progression analysis.
Deeper dive: How AI is advancing OCT analysis in glaucoma
Dr. Radcliffe discussed several ways that AI can support ophthalmologists in their day-to-day workflows, but one of the most interesting points he raised was the technology’s ability to support visual field and OCT analysis in glaucoma.
Glaucoma is a complicated disease, and doctors cannot diagnose it by looking at a single image or test result.
Instead, they must integrate multiple sources of clinical information, including:
- What a patient’s eyes look like
- Retinal nerve fiber layer (RNFL) thickness
- Optic nerve appearance
- Optical coherence tomography (OCT) findings
Ophthalmologists also need to consider functional data, such as visual field testing, visual field progression over time, and the severity of visual field loss. Because so many factors go into making a glaucoma diagnosis, researchers have found it difficult to adequately train AI algorithms.1
The challenges of training AI for glaucoma diagnosis
To effectively train AI, researchers need to give it a ground truth––a reliable reference standard or agreed-upon classification that everyone in the field can agree upon.2
The Collaborative Community for Ophthalmic Imaging’s Glaucoma Workgroup began addressing this challenge during virtual meetings held in 2020 and 2022. However, workgroup members noted that definitions of glaucoma are often subjective and that different reference standards are often required depending on the clinical setting.3
At the same time, most of the currently available data sets used to train AI on glaucoma were collected in clinical settings. Imaging developed in these scenarios is often high-quality, but it doesn’t always translate to practices that have older equipment or non-specialized technicians taking photos.
These datasets also tend to lack racial and ethnic diversity, which may affect their ability to detect glaucoma across different demographic groups.4 Together, these factors make it challenging to ensure AI performs reliably in real-world clinical settings.
Emerging applications of AI in OCT analysis
Despite these obstacles, AI can support clinical decision-making, particularly when it comes to OCT image analysis.
“You can currently use AI for visual field and OCT analyses, for example, you can take pictures of an OCT or a series of OCTs; just mask the patient name, and you can get analysis on that data—such as visual field progression.”
Machine learning algorithms can analyze large amounts of imaging data and identify patterns that may help ophthalmologists detect signs of glaucoma more accurately and consistently.5
Take myopic glaucoma detection, for example. People with extreme nearsightedness experience eye changes that are very similar to those caused by glaucoma. AI models trained on both sets of data may make it easier to distinguish glaucoma-related changes from those caused by nearsightedness or vice versa.6
AI tools can also make it easier to know if glaucoma is worsening over time. Deep learning analysis of OCT B-scans, visual field test results, intraocular pressure measurements, and blinded patient data may help:
- Predict disease progression
- Identify higher-risk patients
- Personalize monitoring schedules
These capabilities can help ophthalmologists make more informed decisions by identifying patients who require closer monitoring or earlier intervention.
Current limitations of AI in glaucoma care
Although AI shows promise, it cannot independently diagnose or manage glaucoma. It’s also important to remember that patient privacy and HIPAA still apply. If you plan on using these tools in your practice, always remove names and any other identifying information before uploading images or clinical notes.
Furthermore, keep in mind that none of the AI tools currently available have official regulatory approval. As Dr. Radcliffe explained, "We don't currently have FDA-approved tools that are bespoke for glaucoma that we can plug in right now to our EMRs, OCTs, or visual field databases ... I think those tools will come; it will likely be a little while before they start popping up, and I'm sure once they do, they'll be very helpful."
Final thoughts
AI can be a valuable support tool for ophthalmologists in many ways. It’s fast, efficient, and capable of analyzing large volumes of data, but it won’t be replacing clinicians any time soon; the doctor-patient connection is just too important.
Even so, these tools may help strengthen those relationships and improve practice operations as a whole by enhancing communication, streamlining daily responsibilities, and supporting clinical decision-making. For those interested in incorporating AI into their workflows, early experimentation is an effective way to get more comfortable with the technology and identify everyday use cases.
“The exciting thing is that if you haven't been using AI, you've got a lot of great things to look forward to. If you already have, I hope I gave you even just one idea that would help you in your practice.”
This video was written by Chad Birt based on the recorded video by Dr. Radcliffe.
