If we could get a glimpse into the future, what would clinical practice a decade or two from now look like? For certain, it would include a heavy emphasis on telemedicine that involves some form of artificial intelligence (AI). We’ve just witnessed the first fast-track FDA-approved example of an AI system (IDx-DR, IDx Technologies) to detect diabetic retinopathy, so AI has definitely entered the field of medicine and eye care in a formal fashion. Teams around the world are using AI for imaging diagnostics (image analysis and machine learning tools) to detect early macular degeneration and glaucoma. And I’m certain there’s much more to come.
AI Today and Tomorrow
By any standard, AI is not a recent development. We’ve used AI in its crudest form for many years—most of us just haven’t realized it. Glaucoma management relies on regression analysis from images of the nerve fiber layer and ganglion cell complex. When we interpret visual fields, we receive automated information that helps us determine the best course of action for each patient.
Machine-learning AI also plays an important role in disease diagnosis verification with multiple optical coherence tomography (OCT) image readings and intraocular lens calculations of corneas with significant pathology or after corneal surgery to improve outcomes in cataract surgery patients.1 Machine learning is a sub-field of AI, and deep learning is a sub-division of machine learning.1 Diabetic retinopathy detection currently uses machine-learning algorithms, which will continue to serve us well in other areas in the future.
Once an AI system can recognize patterns or markers of disease, it can be used for automated diagnosis and detection of almost any condition seen in clinical practice.2 AI systems that feature deep learning with conventional neural networks can identify any number of disease features, leading the way to future uses in cornea and contact lens-related technologies.2,3 Using deep learning to learn more about a disease pattern of expression may provide additional markers for diagnosis and for staging and gauging prognosis.4
Surprises in Store
In addition to posterior segment applications, researchers are developing and validating AI systems for the automatic diagnosis and characterization of anterior segment diseases such as (1) grading cataracts in the pediatric/newborn population based on images obtained by slit lamp and (2) assessing keratoconus based on Scheimpflug images that provide measurements of curvature, thickness and opacity.3
An especially exciting initiative in eye care AI comes from a collaboration between the Moorfields Eye Hospital in London and the Google AI team, DeepMind. These teams created an AI system that combines two deep-learning systems with the ability to detect multiple ophthalmic diseases based on an analysis of 3D OCT data by creating a tissue map and then inspecting for potential markers of that disease.3
Some experts worry about AI’s limitations, such as the technology’s ability to evaluate context accurately and the applicability for a system’s use.4 AI systems will need model and performance metrics to assess efficacy and accuracy at the same standard as the rest of us. We’ll still need humans to oversee the machine-driven technologies that will be developed in the near future.
Assistive AI won’t make us more productive but it will allow greater patient access, lower costs and opportunities to learn more about a particular disease or condition for better outcomes possible.2
Artificial intelligence and, in particular, deep learning, will surely change the way we work and will indeed help us serve a broader patient base. AI is just one of the many ways practice will change in the future. As resources advance, disruptive technologies will flourish. It’s difficult to predict how significantly future advances in AI systems will impact the management of anterior segment conditions. The future will be exciting but also poses new challenges to all of us engaged in practice.
1. AI Applications in Ophthalmology Achieve Human Expert-Level Performance. Healio Ocular Surgery News. June 10, 2018. https://www.healio.com/ophthalmology/technology/news/print/ocular-surgery-news/%7B373d8cf8-f72b-43fe-b362-29ba35c416ba%7D/ai-applications-in-ophthalmology-achieve-human-expert-level-performance. Accessed November 15, 2018.
2. Roach L. Artificial intelligence. EyeNet. 2017;77-83.
3. Varma R. How AI benefits patients and physicians. Ophthalmology Times. 2018. http://www.ophthalmologytimes.com/article/how-ai-benefits-patients-and-physicians. Published September 15, 2018. Accessed November 15, 2018.
4. Trikhar, S. Embracing and managing AI in the eye care workplace. Ophthalmology Times Europe. http://europe.ophthalmologytimes.com/ophthalmology/embracing-and-managing-ai-eye-care-workplace.