AI advancements expand oculomics capabilities, fuel collaborative projects
Key takeaways:
- Research for oculomics, the study of the eye as a window to systemic health, has expanded in recent years.
- Diverse data sets are key for the training and performance of AI models in this field.
With AI moving at a fast pace, oculomics — the study of the eye as a window to systemic health — is expanding its potential and gaining increasing attention.
The past few years have seen an abundance of studies and initiatives aimed at creating strategic partnerships for collaborative projects in this area. Oculomics-based AI systems promise to become a transformative tool for risk stratification and early diagnosis of a wide range of systemic conditions, ultimately leading to more equitable access to care.
“We have long been familiar with the concept of associating abnormalities in the eye with systemic conditions, for instance, arteriovenous nicking with hypertension,” Amitha Domalpally, MD, PhD, research director of the Wisconsin Reading Center, University of Wisconsin-Madison, said. “What AI has brought is the ability to identify associations that are beyond human perception, integrating imaging, lab tests, genetics and maybe variables like Apple Watch data.”
Oculomics is going to be an emerging and transformative technology, the living embodiment of the AI impact in ophthalmology, according to Healio | OSN Neurosciences Section Editor Andrew G. Lee, MD, professor of ophthalmology, neurology and neurosurgery at Weill Cornell Medical College.
“It will become a multidisciplinary and interdisciplinary way to answer the question, ‘Is the eye really a window to systemic disease and the central nervous system?’” Lee said. “And it will answer it in a quantitative way rather than qualitative only. It will confirm all sorts of hypotheses that have been generated in the pre-AI and pre-oculomics era. It’s exciting to be a part of it, but with that comes a great responsibility.”
Biological foundation and evolution of oculomics
The eye and brain share a common embryologic origin, Lee said.
“The eye is composed of ectodermal and neuroectodermal derivatives that are similar to the brain’s neuroectoderm, and that shared embryologic origin creates a connection that makes it possible for us, clinically, to look in the back of the eye and have a unique window into the central nervous system. No other part of the body offers this possibility,” he said.
Changes in the blood vessels of the retina might mirror those occurring in the cerebral vessels, revealing the presence of neurological conditions. In addition, because the ocular vasculature connects with the systemic circulation, changes in retinal microvasculature may be signs of systemic conditions such as diabetes and cardiovascular disease.
In the mid-19th century, with the invention of the ophthalmoscope, researchers were able to hypothesize an association between specific retinal microvasculature characteristics, renal disease and hypertension.
“We could only look and see but couldn’t measure anything. Later, with the emergence of digital fundus photography, we were able to transition from qualitative to quantitative evaluation,” Lee said. “A further advancement was OCT, which allows us to have a noninvasive optical biopsy of the retina, the optic nerve and the nerve fiber layer. And now we have OCT angiography for the blood vessels. They show us to the micron level the pathological alterations of the microvasculature, which are a reflection of the same microvascular changes that are occurring elsewhere in the body with diabetes, hypertension and kidney disease and in the brain, for example, with neurodegenerative diseases or stroke.”
Domalpally recalled how, about 25 years ago, researchers at the Wisconsin Reading Center developed the Interactive Vessel Analyzer (IVAN), an algorithm to measure blood vessel diameter from fundus photography. It was used in many public health studies and showed how vessel diameter was related to hypertension, stroke and cardiovascular events.
“This was oculomics already, but it was not AI-driven,” she said. “The second step was when we moved into structure with OCT: We were able to look at retinal layers and predict neurodegenerative diseases like Alzheimer’s and Parkinson’s. But that also doesn’t need AI.”
The third step was when AI came into play, moving the analysis beyond the vessels and structure to unknown, unexplainable ways of recognizing patterns and identifying early signs of diseases that are beyond human perception. In many cases, human observers are still unable to follow and decipher the logical or causal chain leading to that particular output.
“Examples are the ability for AI to look at a retinal photo and determine sex, estimate your hemoglobin level and whether you have anemia, what is your kidney function and how long you have been smoking,” Domalpally said.
“In short, oculomics is an interdisciplinary way of combining the power of ocular imaging technologies with bioinformatics and AI,” Lee said. “The power contained in looking at that data allows us to do way more things than we ever did before.”
Collaborative big data projects
Access to large sets of diverse data is critical to the training and performance of AI models. The United Kingdom has been a leader in collecting big datasets with the UK Biobank project, which since 2006 has been gathering health-related data on half a million volunteer participants.
“The UK Biobank has been the biggest gift to oculomics,” Domalpally said. “Asia also has huge numbers, so they have been able to do a lot of very innovative oculomics studies. In the U.S., there is a lack of those gigantic datasets, but projects are ongoing.”
With her team, Domalpally is involved in the collection and labeling of large datasets for future AI development and in studies that have great potential for oculomics.
“What we need is diverse, multidimensional data,” she said. “As a reading center, we have been part of the NIH Multi-Ethnic Study of Atherosclerosis. A large amount of data is collected on systemic parameters as well as lifestyle factors and psychosocial factors potentially linked to cardiovascular disease, and retinal photography is included in the protocol.”
The study started in 2000 and gathered a diverse multiethnic population of 6,814 men and women without known cardiovascular disease. The aim is to determine the risk factors and biomarkers of disease onset as well as progression from subclinical to clinically overt disease.
The Coronary Artery Risk Development in Young Adults (CARDIA) is another study, led by Manjot K. Gill, MD, MS, a retina specialist at Northwestern University Feinberg School of Medicine, and Donald M. Lloyd-Jones, MD, ScM, a cardiologist and epidemiologist at Boston University.
“At our reading center, we will be collecting retinal images and OCT angiograms for the CARDIA, which includes over 5,000 people equally distributed into one-quarter white, one-quarter Black, one-quarter Hispanic and one-quarter Asian, recruited at the age of 18 to 30 years, with now a follow-up of almost 40 years,” Domalpally said. “Every parameter of their body has been measured, and retinal images are now being collected.”
These studies are creating a rich dataset for oculomics, looking into the aging process, what causes cardiovascular disease and how it can be predicted from retinal images.
“The CARDIA study is also looking into the biological age gap — that is, the difference between chronological and biological age as predicted by the retinal images,” Domalpally said. “This gives us insights into how the body is aging internally in people who are 50 years old chronologically and may be biologically closer to 30 on the inside vs. being 50 years old by age and 70 years old inside.”
The National Health and Nutrition Examination Survey is a national project collecting data on the health and nutrition of adults and children. It is performed nationwide by the CDC, and the Wisconsin Reading Center is collecting and processing the data.
“We collect all the retinal images and measurements, but the CDC is collecting data on health exams and laboratory tests related to every part of the body so that you have a complete profile of each participant. This is another dataset suitable for oculomics-related research,” Domalpally said.
Developing and testing AI models
Another part of Domalpally’ s work consists of developing and testing AI models. She is currently involved in the Carotenoids in Age-Related Eye Disease Study, an ancillary study of the Women’s Health Initiative.
“There are more than 1,000 women who have had their retinal images taken and their nutritional history and disease history collected,” she said. “In collaboration with Dr. Amy Millen, University at Buffalo, we have an NIH grant right now to look into their gut biome. Their stool samples have been collected and analyzed, and we are getting insights into the connection between gut biome, nutrition and the retina. By looking at retinal images, we might be able to tell how good their food intake is and what their microbiome is because that relates to the inflammatory processes that are involved in macular degeneration but also in other diseases and the overall health of the body.”
In its multimodal approach, oculomics also includes genomic data, which can provide unique information on an individual’s risk to develop specific conditions. A $3 million grant has been allocated to the University of Wisconsin to analyze genomic and other data related to oncologic patients, and a subset of these patients have retinal imaging.
“We are exploring the possibility of using retinal images for the diagnosis, prognosis and monitoring of cancer management. This would save a lot of the burden of many other expensive tests,” Domalpally said. “These are just a few examples of all you can do with oculomics. It is as creative as you can get. It is a very open field, and we’re only scratching the surface.”
Multimodal, multiethnic approach
Sally L. Baxter, MD, MSc, associate professor at the University of California San Diego, is also involved in building datasets that are relevant to oculomics. One of these projects is the NIH Bridge2AI program.
“Within the program, we are working at AI-READI, a data-generation project for type 2 diabetes that includes patients across different races and ethnicities,” Baxter said. “We are collecting a whole slew of retinal imaging and also a wide array of systemic data and measurements. The aim is to eventually try to use retinal imaging to understand different severity levels of diabetes and how to move from disease to health. This is what we call salutogenesis.”
A multimodal approach in ophthalmology means combining multiple imaging modalities, but AI-empowered oculomics employs a wider and more complex multimodal approach. The AI-READI dataset gathers the whole array of eye imaging, including color photography, OCT, OCT angiography and fluorescence lifetime imaging ophthalmoscopy, as well as blood testing, cognitive testing, monofilament testing, electrocardiogram, continuous glucose monitoring and more.
“The idea is to weave together the imaging along with all these other markers of different outcomes for diabetes to try to better understand more intricately the links between diabetes and other systemic diseases as well as with the eyes,” Baxter said.
This dataset is broadly available to the wider research community, and anyone with an .edu email address can access the data at https://docs.aireadi.org/.
Another initiative in which Baxter is involved is the NIH All of Us Research Program, aimed at building a nationwide diverse database that will serve as a source for many studies on a variety of health conditions. One of them, already launched in collaboration with the National Eye Institute and the National Institute of Biomedical Imaging and Bioengineering, is the Eyes on Health Partnered Research Study to advance oculomics research. UC San Diego is one of the sites enrolling participants to collect eye imaging.
“One of the core values of All of Us is to welcome people of all backgrounds, from all racial and ethnic groups, to reflect the diversity of our population,” Baxter said.
The Alliance for Healthcare from the Eye is another project in which UC San Diego is involved. It is a collaboration among health care providers, technology developers and policymakers to integrate oculomics into routine clinical practice.
“We are having discussions with Microsoft and Amazon. The idea is to get retinal cameras outside of eye clinics and more in primary care and even retail settings to screen for systemic diseases,” Baxter said.
From research to the real world
Ocular image-based AI has strong potential to become a cost-effective screening tool, particularly for large-scale population health programs, according to Andrzej Grzybowski, MD, PhD, professor of ophthalmology at the University of Warmia and Mazury, Olsztyn, Poland. The main reason is that retinal imaging is relatively fast, noninvasive and inexpensive compared with other diagnostic modalities such as MRI or CT.
“AI-driven automated grading of retinal images can significantly reduce the workload for specialists and allow for earlier detection of vision-threatening and systemic diseases,” Grzybowski said. “Pilot studies in diabetic retinopathy screening have already shown that AI can reduce costs while maintaining high diagnostic accuracy. With further refinement and integration into telemedicine platforms, such tools could be scaled to primary care or pharmacy settings. This would democratize access to advanced diagnostics, especially in low-resource countries.”
In clinical practice, automated AI tools are already assisting in screening programs, especially in underserved areas where ophthalmologists are scarce.
“We can also use AI models to analyze retinal microvasculature as a biomarker of systemic cardiovascular risk,” Grzybowski said. “These approaches allow for earlier referral to specialists such as cardiologists or endocrinologists. In summary, oculomics is gradually moving from research into everyday clinical reality.”
Oculomics will open new pathways for interdisciplinary collaboration and teamwork among different specialties, bridging ophthalmology with cardiology, neurology, endocrinology and even psychiatry.
“For example, research on retinal imaging in Alzheimer’s disease requires collaboration between ophthalmologists and neurologists to validate ocular biomarkers against cognitive decline,” Grzybowski said. “Similarly, joint efforts between ophthalmologists and cardiologists are essential to link retinal vascular patterns with cardiovascular outcomes. These collaborations also extend to data scientists and AI engineers who play a crucial role in developing robust algorithms. Such partnerships not only enhance diagnostic accuracy but also accelerate translation from research to clinical care. This trend could transform the way medicine approaches systemic disease management.”
With the aging of the population, neurodegenerative diseases are becoming more prevalent. It is in this field that oculomics could become a true game changer, according to Baxter.
“More people are struggling with these conditions, but we still don’t have great treatments and good early risk markers to be able to identify and treat them early,” she said. “For cardiovascular disease, we have a lot of other good tests and good lab markers to identify people at risk, but for neurological diseases, there’s a lot more room for improvement and a lot more need for better therapies. Using oculomics and with larger datasets of eye imaging available, we might be able to make gains in that field in the future.”
However, to develop the full potential of oculomics, further research is essential, according to Lee.
“At this initial stage, we’re just confirming what we already knew and hypothesized before: Does what we see correlate with diabetic retinopathy? Does what we see correlate with Alzheimer’s disease and Parkinson’s disease?” he said. “The next phase is doing that same process but with way bigger numbers to make sure the AI isn’t hallucinating a pattern that isn’t there. Our job is to validate that and make sure it’s true.”
Challenges and strategies of a paradigm change
Despite rapid progress, the transition of oculomics to real-world screening programs and clinical practice is not without challenges.
“One limitation is still the scarcity of large, diverse and standardized datasets. AI models trained on one population often fail to generalize to another due to ethnic and demographic differences,” Grzybowski said. “Another issue is the ‘black box’ nature of many deep learning models, which makes clinical validation and regulatory approval more complex. Integration into real-world workflows remains difficult, as health care systems require interoperability with electronic health records.”
There are also ethical issues regarding people’s data and their privacy.
“We need countermeasures before we have widespread screening as a general rule because a problem with big data is that you are surrendering your security and your privacy,” Lee said. “If disease can be predicted, this information might be used in a way that we did not intend.”
One countermeasure is the federated learning approach that is now widely applied to the creation of machine learning models trained on multiple datasets.
“You train the model across several independent institutions without sharing patients’ data,” Domalpally said. “However, it is important to note that none of the currently developed oculomics algorithms are FDA approved, so they cannot be used in patient care.”
To dispel some of the concerns related to making retinal images more widely available to the greater research community, the American Academy of Ophthalmology made an official statement that retinal images are not biometric identifiers the way that fingerprints are.
“There’s been a lot of discussion around this, also as a consequence of popular movies showing retinal scanners that unlock devices, identify criminals or open a secret vault,” Baxter said. “It’s good for the general public to know that retinal images are not considered biometric identifiers the same way that fingerprints or iris images are. At airports, we are using iris images now for biometric identification, like through the CLEAR program, for example, but the iris is a very different structure than the retina, and the general public doesn’t really know that difference.”
Incorporating AI into the health care environment holds great promise but also unknowns, uncertainties and question marks. These should not be barriers, and researchers should embrace them, maintain focus and move forward, according to Lee.
“The baseball player and ‘philosopher’ Yogi Berra said, ‘It’s tough to make predictions, especially about the future,’” he said. “Anything that we would predict about AI and oculomics is going to be uncertain, but one thing that is certain is that change is coming. I don’t know what that change is exactly going to look like, but I think oculomics in the future is going to be an emerging, exciting and transformative technology.”
Click here to read the Point/Counter to this Cover Story.
For more information:
Sally L. Baxter, MD, MSc, of University of California San Diego, can be reached at s1baxter@health.ucsd.edu.
Amitha Domalpally, MD, PhD, of Wisconsin Reading Center in Madison, Wisconsin, can be reached at domalpally@wisc.edu.
Andrzej Grzybowski, MD, PhD, of University of Warmia and Mazury, Olsztyn, Poland, can be reached at ae.grzybowski@gmail.com.
Andrew G. Lee, MD, of Weill Cornell Medical College, New York, and Blanton Eye Institute, Houston Methodist Hospital, can be reached at aglee@houstonmethodist.org.
Sources/Disclosures
Source:
Healio Interviews
References:
- An S, et al. Prog Retin Eye Res. 2025;doi:10.1016/j.preteyeres.2025.101352.
- An S, et al. Sci Rep. 2024;doi:10.1038/s41598-024-71517-w.
- Arnould L, et al. Ophthalmol Ther. 2023;doi:10.1007/s40123-022-00641-5.
- Bisen JB, et al. J Cardiovasc Dev Dis. 2025;doi:10.3390/jcdd12060230.
- Cabrera DeBuc D, et al. J Clin Med. 2019;doi:10.3390/jcm8070996.
- Chaitanuwong P, et al. Neurol Ther. 2023;doi:10.1007/s40120-023-00526-0.
- Chan YK, et al. Taiwan J Ophthalmol. 2023;doi:10.4103/tjo.TJO-D-23-00018.
- Choi JY, et al. EPMA J. 2024;doi:10.1007/s13167-024-00378-0.
- DeBuc DC. Lancet Digit Health. 2023;doi:10.1016/S2589-7500(23)00047-X.
- Denniston AK, et al. Curr Opin Ophthalmol. 2022;doi:10.1097/ICU.0000000000000887.
- Domalpally A, et al. Lancet Digit Health. 2021;doi:10.1016/S2589-7500(21)00140-0.
- Ghenciu LA, et al. Biomedicines. 2024;doi:10.3390/biomedicines12092150.
- Gim N, et al. Exp Eye Res. 2025;doi:10.1016/j.exer.2025.110341.
- Grzybowski A. Artificial Intelligence in Ophthalmology. Springer; 2025.
- Grzybowski A, et al. OCT and Imaging in Central Nervous System Diseases. 3rd ed. Springer; 2025.
- Huang Y, et al. EPMA J. 2023;doi:10.1007/s13167-023-00315-7.
- Kermany DS, et al. Cell. 2018;doi:10.1016/j.cell.2018.02.010.
- Kim BR, et al. EPMA J. 2022;doi:10.1007/s13167-022-00292-3.
- Kim HK, et al. Int Ophthalmol. 2025;doi:10.1007/s10792-025-03500-x.
- Li C, et al. Int J Surg. 2025;doi:10.1097/JS9.0000000000002236.
- Li H, et al. Healthcare (Basel). 2023;doi:10.3390/healthcare11121739.
- Li Z, et al. J Adv Res. 2025;doi:10.1016/j.jare.2024.11.018.
- Majeed A, et al. Front Aging Neurosci. 2021;doi:10.3389/fnagi.2021.720167.
- Ong J, et al. Asia Pac J Ophthalmol (Phila). 2024;doi:10.1016/j.apjo.2024.100095.
- Patterson EJ, et al. Ophthalmol Ther. 2024;doi:10.1007/s40123-024-00942-x.
- Silverstein SM, et al. Schizophr Bull. 2021;doi:10.1093/schbul/sbab011.
- Suh A, et al. Ann Biomed Eng. 2023;doi:10.1007/s10439-023-03365-0.
- Suh A, et al. Eye (Lond). 2024;doi:10.1038/s41433-024-03132-y.
- Wagner SK, et al. BMJ Open. 2022;doi:10.1136/bmjopen-2021-058552.
- Wagner SK, et al. EClinicalMedicine. 2024;doi:10.1016/j.eclinm.2024.102493.
- Wagner SK, et al. Transl Vis Sci Technol. 2020;doi:10.1167/tvst.9.2.6.
- Wang J, et al. Theranostics. 2025;doi:10.7150/thno.100786.
- Weinreb RN, et al. JAMA Ophthalmol. 2025;doi:10.1001/jamaophthalmol.2025.0881.
- Wu JH, et al. J Clin Med. 2022;doi:10.3390/jcm12010152.
- Zhang Y, et al. BioData Min. 2024;doi:10.1186/s13040-024-00363-3.
- Zhu Z, et al. Prog Retin Eye Res. 2025;doi:10.1016/j.preteyeres.2025.101350.
Disclosures: Baxter, Domalpally, Grzybowski and Lee report no relevant financial disclosures.
Ask a clinical question and tap into Healio AI's knowledge base.
- PubMed, enrolling/recruiting trials, guidelines
- Clinical Guidance, Healio CME, FDA news
- Healio's exclusive daily news coverage of clinical data
Read more about
