Michael B. Rivers, MD, is the director of EMA Ophthalmology at Modernizing Medicine.

Ophthalmology is facing an impending workforce crisis: Not only do more ophthalmologists retire each year than enter the field (about 550 vs 450), but aging populations and a growing prevalence of diabetes means an increasing demand for eye care. With these converging pressures, technologies are uniquely poised to help — if we let them.

Among those in ophthalmology, there is still hesitancy surrounding technology like artificial intelligence (AI) and deep learning (DL). Most hesitation is rooted in the idea that AI and DL are meant to replace or reduce physician support in critical areas of patient care. When speaking with providers and peers about these innovative tools, something I reiterate is that AI is not replacing ophthalmologists any time soon; instead, it’s giving us a chance to help automate redundant, time-consuming tasks while freeing us up to more fully focus on our patients. Why should we spend hours on end analyzing patient tests when we can leverage technology that helps save time and produces reliable results to help patients receive more timely treatment? Here are a few ways technology aids ophthalmologists now, as well as a look toward future innovation.

The Current State

According to the US Centers for Disease Control and Prevention, approximately 12 million people 40 years of age and older in the United States have vision impairment.¹ Diabetic retinopathy is a major contributor, with one-third of adults over the age of 40 suffering from the condition.² For this reason, primary care physicians are more frequently encouraging patients — especially those with diabetes — to build eye exams into their yearly health care routine. At the same time, with more ophthalmologists retiring and fewer entering the field each year, resources to perform these additional screenings are tight, placing even more responsibilities on ophthalmologists and contributing to burnout.

This is an area where AI can help. As demonstrated by the Google Brain initiative, DL and self-optimizing algorithms can inspect mass amounts of fundus photographs with a high degree of accuracy to detect the presence and severity of diabetic retinopathy and help track the progression of the disease.³ In fact, in April 2018, the FDA approved the first AI-based diagnostic tool for detecting diabetic retinopathy.⁴ It uses a self-guided fundus camera to take high-resolution photos of the patient’s eyes then scans for signs of the disease. Having the ability to streamline and automate eye screenings during patient visits using AI and DL can make a major difference when it comes to efficiently finding those patients that need an examination by an ophthalmologist.

As we become more comfortable working alongside AI, we open up new avenues to make a difference in patients’ lives. A recent study published by the American Academy of Ophthalmology found that when it comes to diagnosing diabetic retinopathy, AI and physicians are much more effective working in tandem than either entity working alone.⁵ This study reiterates the fact that while this technology has incredible potential, we are not at a point where we risk replacing trained physicians. We need to understand that embracing technology can revolutionize and democratize patient care as it now stands.

The Promising Potential

As we look toward the future of AI and DL applications, telemedicine holds great promise for early disease detection, and it could help retina specialists improve public health overall by making eye care even more accessible. Innovative, intuitive technology like DL can help expand where and how often eye screenings take place, making the process more efficient for ophthalmologists and hopefully providing patients the treatment they need sooner.

For example, telescreening solutions could allow patients to be tested for diabetic retinopathy and macular degeneration far more frequently and conveniently using portable cameras and AI assistance for diagnosis. Images could be taken by primary care physicians at the yearly physical or, in the future, by patients themselves at a health station in their local pharmacy or even in their home.

In fact, tests of this kind exist that are showing promising results. One company recently used telescreening and AI analysis to scan 1,674 patients for diabetic retinopathy (EyeArt; Eyenuk).⁶ The technology correctly determined that 480 patients were positive for the disease, and out of the approximately 1,200 patients the technology marked as not in need of a referral, only 14 were confirmed as referral-warranted by the center analyzing the results. Those results will likely improve as companies continue to refine and enhance these tools to meet the needs of providers and patients alike.

In the future, AI could be used to advance OCT tests by more accurately scanning images and analyzing granular details too small for the human eye to detect. Going further, the algorithm could leverage social determinants of health and other data from the patient’s electronic health record to determine the patient’s risk factors for disease to make more accurate diagnosis.

Overall, AI and DL applications are showing promise across the board in detecting more than 50 types of eye diseases, and it may only be a few years before patients are able to snap a photo for screening at their local pharmacy and upload for an initial screening by AI.⁷

The Future of AI in Ophthalmology

There is no denying that AI and DL will continue to influence ophthalmology. Applications and advancements of these seemingly futuristic technologies have driven enhanced diagnosis and detection of diseases like diabetic retinopathy substantially, helping us improve public health while increasing efficiency for providers. While we are still in the infancy phase of truly grasping its potential, it’s clear to see that embracing this technology will only make retina specialists better, more efficient care providers.

References

1. Centers for Disease Control and Prevention. Vision Health Initiative Fast Facts. https://www.cdc.gov/visionhealth/basics/ced/fastfacts.htm . Accessed February 21, 2020.
2. Centers for Disease Control and Prevention. Diabetic Retinopathy. https://www.cdc.gov/visionhealth/pdf/factsheet.pdf . Accessed February 21, 2020.
3. Gulshan V, Peng L, Coram M, et al. Development and validation of a deep learning algorithm for detection of diabetic retinopathy in retinal fundus photographs. JAMA. 2016 Dec 13;316(22):2402-2410.
4. US Food and Drug Administration. FDA permits marketing of artificial intelligence-based device to detect certain diabetes-related eye problems. News release. https://www.fda.gov/news-events/press-announcements/fda-permits-marketing-artificial-intelligence-based-device-detect-certain-diabetes-related-eye . Accessed February 21, 2020.
5. American Academy of Ophthalmology. Artificial intelligence can support ophthalmologists, not replace them. https://www.aao.org/eye-health/news/artificial-intelligence-diabetic-retinopathy-diagn . Accessed February 21, 2020.
6. Lim J, Bhaskaranand M, Ramachandra C, Bhat S, Solanki K, Sadda S. Artificial intelligence screening for diabetic retinopathy: analysis from a pivotal multi-center prospective clinical trial. Presented at: ARVO Imaging in the Eye Conference 2019. Vancouver, BC, Canada, 2019.
7. De Fauw J, Ledsam JR, Romera-Paredes B, et al. Clinically applicable deep learning for diagnosis and referral in retinal disease. Nat Med. 2018;24(9):1342-1350.