“The future is already here - it’s just not very evenly distributed.” This well-known quote by William Gibson, the renowned science fiction writer who foresaw the rise of the internet and coined the word cyber, perfectly captures the state of genetics and artificial intelligence in ophthalmology today.
Inherited retinal diseases (IRDs) are rare genetic eye disorders causing progressive vision loss and often blindness, resulting from mutations in genes essential for retinal function. These conditions are significant because they are a leading cause of incurable blindness in younger individuals, profoundly impacting quality of life. Collectively, IRDs represent a substantial public health burden due to their lifelong impact and the need for specialized care. Understanding the genetic basis of these conditions is crucial for accurate diagnosis, genetic counseling, and predicting disease progression.
The rapidly advancing field of IRD research, with innovative gene therapies and targeted treatments, highlights the importance of timely and accurate genetic diagnoses for patient eligibility. However, diagnosing IRDs can be complex, often requiring specialized expertise and genetic testing only available in very few centres around the world, leading to diagnostic delays and hindering access to care. Consequently, outside of major hospital centres IRD diagnosis is still a slow, fragmented process. Even in the UK, a country with an advanced genomic medicine service, patients can wait years for a confirmed genetic diagnosis, missing windows of eligibility for trials or emerging therapies. Although genetic testing is without a doubt the technology of the future, it is still not available in most countries for clinical care. Furthermore, to promote the adoption and leverage the utility of genetic testing for the benefit of ophthalmology, the data it generates needs to be linked to clinical data such as retinal imaging.
Over the past six years, we have built Eye2Gene, an ophthalmology platform and database that links genetic data to multimodal retinal scans that include optical coherence tomography (OCT), fundus autofluorescence, and infrared. This resource has allowed us to train and validate the Eye2Gene AI on tens of thousands of images from patients with genetically confirmed diagnoses from specialist centres, in order to predict the genetic cause of IRDs directly from retinal scans in patients with an unknown diagnosis. Eye2Gene has reached performance levels superior to world-leading IRD experts when assessed on fundus autofluorescence scans (30% for specialists vs 76% for Eye2Gene). If deployed across retinal scanners, this platform has the potential to significantly accelerate the patient pathway so that patients from all over the world can have access to an expert IRD service. The Eye2Gene algorithm is now published in Nature Machine Intelligence and integrated into clinical imaging platforms like Heidelberg Engineering’s HEYEX 2 as a research tool, as well as a standalone online platform.
However, AI tools on their own, no matter how sophisticated, are dependent on an entire ecosystem so they can be safely and effectively used.
It’s only through multidisciplinary collaboration between patients, eye care professionals, geneticists, statisticians, health psychologists, computer scientists and software engineers that tools like Eye2Gene can become clinically relevant. Building trust takes time and it starts with co-development – by having clinicians, patients and user researchers embedded in our AI teams – but continues by building a growing evidence base of how AI predictions can improve diagnostic yield, speed up genetic referrals, and help non-experts make confident decisions. This culminates in a certified regulation as a medical device, which can then be integrated into routine care systems and used at scale by trained users.
There is also the wider stakeholder ecosystem in which the technology is deployed. For example, regulatory agencies, reimbursement authorities, genetic testing providers, pharma companies and patient groups are all critical players to the relevance of AI solutions, such as Eye2Gene, and need to be actively engaged and incentivised to ensure mutual success.
Equitable IRD care is now our next frontier. If AI systems are to truly improve global healthcare, we must ensure their benefits are equitably distributed, otherwise we risk deepening existing health disparities. Crucially, these systems need to be able to function in under-resourced clinics, irrespective of geography, where access to IRD specialists is the most restricted. And most importantly, they need to be democratised for patients, who deserve timely, accurate diagnoses regardless of their location. By globally mapping IRD patients around the world with Eye2Gene, we hope to extend the reach of genetic testing, increase health data diversity, accelerate drug development pipelines and incentivise pharmaceutical companies to serve a truly global population.
However, this vision will not be realized if we limit AI development and deployment to countries with existing data and cloud infrastructure. To ensure broader participation and benefit, we must sustainably invest in building new infrastructure, such as data ingestion pipelines, in regions where it is currently lacking, so that they too can help contribute and benefit from the AI systems of tomorrow.
The future of AI in ophthalmology is not just a technological challenge – it’s a human one where we need to work together to increase the size of the pie so that it can feed everyone, everywhere. Not only the select few.
Inherited retinal diseases (IRDs) are rare genetic eye disorders causing progressive vision loss and often blindness, resulting from mutations in genes essential for retinal function. These conditions are significant because they are a leading cause of incurable blindness in younger individuals, profoundly impacting quality of life. Collectively, IRDs represent a substantial public health burden due to their lifelong impact and the need for specialized care. Understanding the genetic basis of these conditions is crucial for accurate diagnosis, genetic counseling, and predicting disease progression.
The rapidly advancing field of IRD research, with innovative gene therapies and targeted treatments, highlights the importance of timely and accurate genetic diagnoses for patient eligibility. However, diagnosing IRDs can be complex, often requiring specialized expertise and genetic testing only available in very few centres around the world, leading to diagnostic delays and hindering access to care. Consequently, outside of major hospital centres IRD diagnosis is still a slow, fragmented process. Even in the UK, a country with an advanced genomic medicine service, patients can wait years for a confirmed genetic diagnosis, missing windows of eligibility for trials or emerging therapies. Although genetic testing is without a doubt the technology of the future, it is still not available in most countries for clinical care. Furthermore, to promote the adoption and leverage the utility of genetic testing for the benefit of ophthalmology, the data it generates needs to be linked to clinical data such as retinal imaging.
Over the past six years, we have built Eye2Gene, an ophthalmology platform and database that links genetic data to multimodal retinal scans that include optical coherence tomography (OCT), fundus autofluorescence, and infrared. This resource has allowed us to train and validate the Eye2Gene AI on tens of thousands of images from patients with genetically confirmed diagnoses from specialist centres, in order to predict the genetic cause of IRDs directly from retinal scans in patients with an unknown diagnosis. Eye2Gene has reached performance levels superior to world-leading IRD experts when assessed on fundus autofluorescence scans (30% for specialists vs 76% for Eye2Gene). If deployed across retinal scanners, this platform has the potential to significantly accelerate the patient pathway so that patients from all over the world can have access to an expert IRD service. The Eye2Gene algorithm is now published in Nature Machine Intelligence and integrated into clinical imaging platforms like Heidelberg Engineering’s HEYEX 2 as a research tool, as well as a standalone online platform.
However, AI tools on their own, no matter how sophisticated, are dependent on an entire ecosystem so they can be safely and effectively used.
It’s only through multidisciplinary collaboration between patients, eye care professionals, geneticists, statisticians, health psychologists, computer scientists and software engineers that tools like Eye2Gene can become clinically relevant. Building trust takes time and it starts with co-development – by having clinicians, patients and user researchers embedded in our AI teams – but continues by building a growing evidence base of how AI predictions can improve diagnostic yield, speed up genetic referrals, and help non-experts make confident decisions. This culminates in a certified regulation as a medical device, which can then be integrated into routine care systems and used at scale by trained users.
There is also the wider stakeholder ecosystem in which the technology is deployed. For example, regulatory agencies, reimbursement authorities, genetic testing providers, pharma companies and patient groups are all critical players to the relevance of AI solutions, such as Eye2Gene, and need to be actively engaged and incentivised to ensure mutual success.
Equitable IRD care is now our next frontier. If AI systems are to truly improve global healthcare, we must ensure their benefits are equitably distributed, otherwise we risk deepening existing health disparities. Crucially, these systems need to be able to function in under-resourced clinics, irrespective of geography, where access to IRD specialists is the most restricted. And most importantly, they need to be democratised for patients, who deserve timely, accurate diagnoses regardless of their location. By globally mapping IRD patients around the world with Eye2Gene, we hope to extend the reach of genetic testing, increase health data diversity, accelerate drug development pipelines and incentivise pharmaceutical companies to serve a truly global population.
However, this vision will not be realized if we limit AI development and deployment to countries with existing data and cloud infrastructure. To ensure broader participation and benefit, we must sustainably invest in building new infrastructure, such as data ingestion pipelines, in regions where it is currently lacking, so that they too can help contribute and benefit from the AI systems of tomorrow.
The future of AI in ophthalmology is not just a technological challenge – it’s a human one where we need to work together to increase the size of the pie so that it can feed everyone, everywhere. Not only the select few.
