Clinical trial shows “eye care copilot” can improve retinal disease diagnoses and subspecialty referral

Clinicians and researchers at The Ohio State University College of Medicine participated in a clinical trial to test the real-world role of an innovative eye care foundation model called “EyeFM.” EyeFM is an artificial intelligence (AI) tool that was pretrained on 14.5 million retinal images taken from five imaging modalities and paired with clinical information from around the world. The main goal was to gauge how EyeFM can be used as an eye care copilot tool to improve the diagnostic performance of ophthalmologists and affect patient outcomes.  

Sayoko E. Moroi, MD, PhD, professor of Ophthalmology, the William H. Havener, MD Endowed Chair and the director of the Havener Eye Institute at The Ohio State University Wexner Medical Center, says her team was invited to participate in “An Eye Care Foundation Model for Clinical Assistance: A Randomized Controlled Trial,” by Ping Zhang, PhD. Dr. Zhang is associate professor of Biomedical Informatics in the College of Medicine and also leads the Artificial Intelligence in Medicine Laboratory at the university and works with researchers around the world.  

Six Ohio State clinicians were invited to participate as opthalmology readers in this clinical trial that randomized 44 ophthalmologists from across North America, Europe, Asia and Africa to determine diagnosis of retinal diseases from retinal images. A new set of retinal images were taken from 668 participants. After study criteria were applied to them, these new images were de-identified and sent to the ophthalmology readers for review. The study compared the retinal diagnoses made by the clinicians as standard care to those randomized to use the AI EyeFM copilot. 

“The results indicated that ophthalmologists with EyeFM copilot achieved a higher correct diagnostic rate, 92% versus 75%, and a higher referral rate, 92% versus 80%, than those receiving standard care,” Dr. Moroi says. 

The study authors recently published these findings in Nature Medicine. The study reveals how EyeFM copilot can be a tool to address needed improvements in managing the bottleneck of patients trying to schedule appointments for dilated eye exams and detection of eye diseases. Furthermore, this approach can increase access for patients who should have annual eye exams, especially for those at high-risk of developing eye diseases.  

When asked about the next steps in implementing these findings in clinical care, Dr. Moroi says the plan is to work toward placing the retina cameras in primary and specialty care clinics where patients visit regularly.   

The team will start small, using an ophthalmology startup company called Optain to place user-friendly cameras which have been Food and Drug Administration-approved, in primary care or specialist offices. They will take pictures of the patients’ retinas without the need for eye drops for dilation. Their retinal images will be assessed for conditions such as diabetic retinopathy, glaucoma, age-related macular degeneration and other vascular problems seen in the retina.

“It’s important to frame this leading-edge technology as a combination of clinician plus machine, as a tool to augment the clinician’s expertise to improve patient care,” she says. “Adoption of such technology will help clinicians gather important clues that lead to understanding, detection and interventions for an individual patient based on the health of their retina, which is an important indicator of the patient’s overall health.”

Dr. Moroi envisions the AI technology use growing to the point where clinicians will review images taken during primary care visits and then reach out to patients for any required follow-up care by a vision specialist. 

“The results could be, ‘Your retina looks fine, let's do this again in a year to screen for retinal conditions,’” Dr. Moroi says. “Or, ‘There are some concerning findings that need to be addressed by an eye specialist.’”

However, Dr. Moroi cautions that AI-based screening of retinal images will not substitute for prescribing glasses for correcting vision, measuring eye pressure as part of the care for glaucoma, or other in-person testing or measurements in the eye specialist’s office. 

AI technologies like EyeFM can aid the analysis of vast retinal imaging datasets to identify subtle patterns and anomalies that a view by the human eye alone might miss during a direct eye exam. Yet, Dr. Moroi says that while AI augments medical expertise, it cannot replace human care. It does provide innovative tools to address the evolving health challenges of the growing aging population, where there is an increasing need for efficient and accurate clinical assessments. 

“The technology is here, and this trial shows its accuracy in application for retina image screening,” Dr. Moroi says. “It has the ability to help those clinicians who work in many areas of care to prioritize their patients who need eye care.”