AI Eye Imaging Models Accelerate Ophthalmic Clinical Trial Recruitment and Drug Development
核心洞察
A University of Colorado Anschutz (搜索) research team developed an AI tool that matches electronic health record data with retinal imaging to rapidly identify eligible patients for ophthalmic clinical trials.
The tool helped researchers recruit five geographic atrophy (搜索) patients in weeks, compared with only five patients across three studies over five years using conventional methods.
A separate AI model, OCTCube-M (搜索), trained on over 1.62 million retinal images, outperformed 2D approaches in detecting retinal diseases and predicting geographic atrophy (搜索) progression.
Clinical trials represent the final, crucial gateway to regulatory approval for new medical treatments, yet ophthalmology studies are notoriously difficult to run because researchers frequently struggle to recruit the minimum number of patients needed. Now, two complementary artificial intelligence (AI) approaches are showing promise for transforming patient recruitment and trial design in retinal disease research.
A group of University of Colorado Anschutz (搜索) Department of Ophthalmology researchers who study retinal disorders has created a new AI tool designed to quickly find and reach patients who qualify for their clinical trials based on retinal imaging. Patients with these disorders must meet stringent imaging criteria to participate in clinical trials, and the AI tool could save researchers immense amounts of time spent searching for the right patients, according to Niranjan Manoharan, MD, associate professor of ophthalmology in the CU Anschutz School of Medicine.
"We basically looked for a way to find clinical data in the electronic health record to see who meets our criteria and then marry that data with imaging AI tools that our group has built," Manoharan said. "Now, we can find these diamonds in the rough."
The Recruitment Challenge in Ophthalmology
Worldwide, patient recruitment presents a major challenge for ophthalmology researchers for numerous reasons. It often takes specialized equipment and diagnostics to detect ophthalmic conditions, and these are not always easy to access. Additionally, patients with a specific type of eye condition may be spread across different geographic areas, or they may have visual impairments that make it harder for them to navigate transportation, attend appointments, and physically participate in a clinical trial.
For Manoharan and his fellow researchers, having specific criteria for patients to meet makes recruitment even harder. He noted that it is challenging for primary health care providers to keep such a long list of criteria in mind and evaluate patients on the fly to see if they might qualify for one of the group's trials.
"Eventually, these kinds of studies fall off the top of people's radars," said Manoharan, who also serves as director of informatics at the Sue Anschutz-Rodgers Eye Center.
Clinical trials in general require a minimum number of patients in order to assess whether a treatment has a measurable, statistically significant effect. There are also typically strict regulations requiring new drugs to be tested on a minimum number of patients. For a new drug to be approved in the U.S., scientists need to test the drug on anywhere from 250 to 1,000 patients, depending on what phase the clinical trial is in. It can take years to find that many patients for ophthalmology studies.
From Five Patients in Five Years to Five Patients in Weeks
Manoharan has been the principal investigator on several patient studies of geographic atrophy (搜索), a condition that occurs in the later stages of dry age-related macular degeneration (搜索) (AMD). In AMD, the macula — a specialized part of the retina — starts to deteriorate, which gradually causes effects such as blurred vision and difficulty reading. With no cure and limited treatment options, AMD is the top cause of vision loss in people over age 50.
Using conventional recruitment methods, Manoharan and fellow researchers were only able to recruit five patients across three studies over the course of five years, and they eventually paused the studies. Now, armed with the new AI tool and updated screening process, the researchers are trying another geographic atrophy (搜索) study and have recruited five patients and counting in a matter of a few weeks.
"We started contacting patients identified by the new AI tool and almost immediately reached our recruitment goal," Manoharan said. "We were very excited."
The tool was developed by a large research team, including Manoharan, led by Jayashree Kalpathy-Cramer, PhD, professor and chief of the CU Anschutz Division of Artificial Medical Intelligence in Ophthalmology. The team received funding and commercial project development support through the CU Anschutz SPARK program, managed by CU Anschutz Innovations.
OCTCube-M: Extracting Insights from 3D Eye Scans
A separate experimental AI model called OCTCube-M (搜索) could help improve patient selection, predict disease progression, and streamline clinical trials by extracting new insights from three-dimensional optical coherence tomography (OCT) eye scans. In a study published in Nature Biomedical Engineering, the model outperformed older 2D AI approaches in detecting multiple retinal diseases, including age-related macular degeneration (搜索) and diabetic retinopathy (搜索). It also improved predictions of disease progression in geographic atrophy (搜索).
Developed using more than 1.62 million retinal images from 26,000 OCT scans, OCTCube-M (搜索) is designed to extract information from the full 3D structure of the eye rather than relying on individual image slices. This approach provides an advantage for identifying patients, measuring disease progression, and improving clinical study design.
One of the most promising applications is clinical trial optimization. AI-based analysis could help identify eligible participants more efficiently, create more uniform study populations, and reduce variability that can affect trial outcomes. Current enrollment strategies often depend on clinical assessments and restrictive inclusion criteria, which can slow recruitment and limit access to eligible patients.
Aaron Lee, M.D., head of the Hardesty Department of Ophthalmology and Visual Sciences at Washington University School of Medicine in St. Louis, said the technology could also help reduce the sample size and time it typically takes to run a study. In collaboration with Genentech, researchers evaluated OCTCube-M (搜索) using randomized controlled trial data and found that the model could be used to significantly reduce the number of patients needed for trials while achieving equal results.
Digital Twins and the Broader "Oculomics" Frontier
Researchers are also exploring whether AI models like OCTCube-M (搜索) could support the development of digital twins — virtual patient models that predict how an individual would likely progress without receiving an experimental therapy. These models could eventually provide additional evidence in clinical trials by comparing a patient's actual outcome with a personalized prediction of disease progression under standard care.
The FDA has shown increasing interest in digital twin approaches, although their use in clinical trials will likely require additional validation. Lee suggested that early applications may be most appropriate in phase 2 studies, where sponsors make decisions about whether to advance therapies into larger and more expensive trials.
Beyond ophthalmology, the technology supports the emerging field of "oculomics," which uses eye imaging to identify signs of systemic disease. Because retinal blood vessels reflect changes occurring throughout the body, AI analysis of OCT scans could potentially help predict conditions such as cardiovascular disease, diabetes, and kidney dysfunction.
Accelerating the Path to Approval
Manoharan said the CU Anschutz group is currently piloting the technology with several external partners and working on commercializing the AI product with support from Innovations. Looking forward, he is hopeful that the AI tool can speed up ophthalmic research by making it much easier to recruit patients for trials. Faster enrollment brings sight-saving therapies to patients sooner and spares them failed screening visits. The development costs it saves can ultimately be reflected in what patients pay.
"If we can accelerate drug approval timelines by months or years, we can get a lot of these drugs to the finish line much sooner," Manoharan said. "If you're spending two or three years on a study only to not recruit enough patients, that costs everybody a lot of money. With this, not only are we saving costs, but we're also hopefully getting drugs approved much faster to treat patients."