Electronic Health Reporter

By Rinat Akhmetov, product lead and ML solutions architect, Provectus.

The use of artificial intelligence (AI) is growing across all sectors, and healthcare is no exception. In fact, AI is particularly well-suited to healthcare applications due to the vast amount of data — from electronic health records (EHR) and clinical trials, to disease registries and claims — that is generated in the industry on a daily basis. 

Ophthalmology is one area where the application of AI technology is more than justified. Faster and more accurate, at-scale eye screening can help diagnose and prevent such eye conditions as amblyopia, strabismus, diabetic retinopathy, glaucoma, age-related macular degeneration, and many others. AI holds the potential to improve patient diagnosis, reduce cost per screening, and expand the availability of eye screening to all.

This article explores how AI can be used in ophthalmology. We will consider the benefits and challenges of AI, outline prospective use cases, and offer a framework for adopting AI.

Ophthalmology is ready for AI innovation

Artificial intelligence is beginning to be used in ophthalmology for a reason. 

A 2020 study researching the use of AI to screen for diabetic retinopathy, a leading cause of blindness, found that AI was able to achieve an accuracy of around 95%, which is comparable to that of expert human graders. Another study used AI to detect glaucoma, also a leading cause of blindness. The AI system was able to achieve an accuracy of over 90% in detecting the disease.

These studies show that the amount of real-world data is enough to develop highly accurate algorithms that can detect disease as well or even better than humans — in all types of eye screens, and at a speed and scale that exceed human potential many times over. 

Given the global shortage of ophthalmologists and optometrists, and the widespread availability of technology (from ready-to-use algorithms to cloud computing), introducing AI to augment the work of  ophthalmologists seems like a sensible solution.

Despite benefits, AI remains a challenge

The potential benefits of using AI in ophthalmology are significant. The improved accuracy and scale of disease detection lead to earlier diagnosis and treatment, which improves patient outcomes. Automated disease screening frees up time for ophthalmologists to focus on other tasks.

However, there are also some challenges associated with using AI. 

AI requires high-quality data for training. And while the volume of data is usually not a problem, finding the right talent to prepare it can be problematic. Only professional ophthalmologists are qualified to label training data in a manner that ensures high accuracy on real-world data in production.

There are risks of false positives or false negatives. Some diseases may be incorrectly diagnosed, while others may be missed altogether. Hence, the importance of prepped data, an infrastructure for AI monitoring and re-training, and human-in-the-loop (HITL) for processing user feedback.

Thankfully, AI technologies are developing so quickly that it becomes easier for practitioners to build eye screening applications from scratch, using open-source tools and cloud services from AWS, Google, or Microsoft.

Practical applications of AI in ophthalmology

There are a number of ways in which AI can be used for disease screening in ophthalmology. 

One example is fundus photography, which is a type of medical imaging that captures an image of the back of the eye. For instance, AI can help capture and interpret the retinal vasculature, to determine risk or presence of diabetes. Similarly, AI can preemptively reveal pathologies that cause blindness and vision loss by enabling at-scale screening for fundus and retina abnormalities at birth.

Another example is the use of Optical Coherence Tomography (OCT). This is a non-invasive imaging technique that uses light waves to take pictures of the retina. These pictures are processed and analyzed by AI to detect any signs of anomalies associated with disease. 

AI can also be used to augment photoscreening applications. GoCheck Kids, a company assisting primary care networks, implements cost-effective pediatric vision screening, and utilizes AI to supplement image analysis and improve user actions, to help ophthalmologists capture the best image possible for further analysis.

The paradigm for AI adoption in ophthalmology

The power of AI lies in its ability to identify patterns and anomalies in data that may be difficult for humans to spot. Nowhere is this more apparent than in the field of ophthalmology, where AI is used for disease screening — detecting anomalous parts of eye screens that may indicate a specific eye condition.

For AI in ophthalmology to work effectively, however, certain conditions must be met.

1. Any disease screening system or application has to have an image labeling component. AI is a work in progress, a system that evolves over time on new data, and users should be able to label new screens and verify low accuracy screens that were previously taken.
2. End-to-end infrastructure for AI has to be in place so that models can be built, trained, deployed, monitored, re-trained, and fine-tuned. Any types of data or model drift, or bias, should be monitored and countered by cyclic model updates.
3. It is better for the solution to live in the cloud. It helps realize such benefits as automatic scalability, high flexibility, and reduced IT costs. It also ensures collaboration efficiency and business continuity. For instance, an eye screen taken with an app by an optometrist in Chicago can be labeled by a highly trained ophthalmologist in LA, with both of them contributing to the improvement of the application’s AI.
4. Having the right UI matters. Doctors taking eye screens should have access to a section that explains why AI made certain decisions, to better understand the signs of detected abnormalities. The labelers should be able to sort existing screens, and markup and feed new screens to the system. A customer-centric UI ensures that doctors do not have to spend time examining screens with no signs of pathology, so they can focus on patients who need assistance.

Conclusion

The potential of AI in healthcare is immense. From streamlining administrative tasks to providing insights for clinical decision-making, AI can help to improve patient outcomes, increase productivity and efficiency of care delivery, and make it easier for wider categories of the population to access healthcare services.

In ophthalmology, AI-powered disease screening is the future. By automating pattern identification, AI can help to increase accuracy while saving time. It can identify individuals who are at risk of developing, or who already have a certain disease, as well or better than human doctors.

It is estimated that by 2050, over 1.8 billion people will suffer from some form of vision impairment. This number could be reduced drastically if preventable vision loss was detected and treated early on. The way forward is to scale disease screening with AI, to enable doctors to focus on patient care while leaving routine work to the machine.