In August 2018, the journal Nature Medicine published a paper describing exciting new research on applications of artificial intelligence (AI) to eye care. The research was conducted by staff members of Moorfields Eye Hospital in London and UCL, a subsidiary of Google’s DeepMind.

The researchers used deep learning to develop an algorithm-focused software program. The program was able to efficiently and accurately identify dozens of eye diseases based on three-dimensional scans. The AI’s diagnostic performance was equal to two of the most accomplished retina specialists in the world. It achieved an enviable 5.5 percent error rate and did not overlook even one urgent case.

Machine learning is the ability of computer programs to "learn" in the absence of specific programming directions. Deep learning is when neural networks can program themselves to perform a wide range of complex tasks based on large data sets. In the case of this research, the deep learning component allows the algorithm to discern patterns in data based on the scans. This led to appropriate referrals to human ophthalmologists who specialize in the eye conditions indicated.

However, these researchers are not the only ones seeking to put artificial intelligence to work on behalf of ophthalmologists. According to the American Academy of Ophthalmology (AAO), these technologies have begun to revolutionize not only science in general, but the everyday practice of ophthalmology. Here are four ways AI is transforming this medical specialty.


1. AI is gaining diagnostic credibility.

At its Mid-Year Forum in 2018, the AAO hosted a session dealing specifically with the use of AI in its field. Among the topics discussed were how AI can use photographs of the fundus — the eye’s inner surface that includes the retina — to aid in diagnosing eye conditions associated with diabetes, such as retinopathy and macular edema.



Participants noted that AI has the potential to offer a higher level of objectivity and precision in analyzing medical images and formulating prognoses than many human physicians.

The AAO observes that artificial intelligence, which began to be developed in the 1950s, has now reached the point at which it can be of significant practical assistance to humans. And thanks to the proliferation of Big Data and increased computing capacity, the efficiency quotient of AI-driven diagnostic tools has increased significantly.


2. AI is on its way to becoming a true medical partner.

The AI pioneered by Moorfields and UCL comes with additional value, according to Nature Medicine. It can offer a cogent explanation of why it made a certain diagnosis, and can point to the precise part of a scan that led to its conclusions.

This capability is a major step forward in the use of AI in medical settings, and allows it to become a more trusted partner in the diagnostic process. The goal of AI diagnostic programs is to analyze large amounts of data and deliver a result. Previous iterations have even achieved this.

Unfortunately, however, these programs were unable to provide the reasoning process and context behind their results. Practicing physicians can appreciate the added dimension that makes the AI behave more like a human colleague who can meet higher-quality standards of care.


3. AI is helping practitioners save time.

InnerEye is a project based at Cambridge University and supported by Microsoft Research. According to an Irish Times article published in August 16, 2018, InnerEye uses AI programs to perform in a matter of seconds the kind of repetitive analytical work that takes skilled practitioners hours to accomplish.


AI programs can quickly assemble a three-dimensional model of an ophthalmologic tumor by marking up the data from hundreds of two-dimensional image scans. Microsoft officials note that InnerEye’s machine learning methods can scan and produce 3-D models of healthy eyes as well. InnerEye uses several algorithms, including Deep Decision Forests.

Expert ophthalmologists and clinicians can extrapolate from AI results as they continue to refine their diagnoses. In each case, the physicians maintain complete control over the process and the results.


4. AI is offering a new way to visit the ophthalmologist.

Another emerging project involves a system called IRIS, or Intelligent Retinal Imaging System. IRIS’s creators describe it as an office visit in a box. The patient’s chin is rested against a supporting strap while a voiceover program guides the process of taking a clear image of the retina within the span of a few minutes.

IRIS’s combined software and hardware apparatus is able to detect diseases such as diabetic retinopathy and glaucoma. As of early 2018, IRIS achieved a 97 percent accuracy rate in its diagnoses. Trained ophthalmologists typically achieve a 92 percent rate of accuracy.

An additional benefit of AI imaging technologies such as IRIS is that images can be saved to the cloud. Images can then be accessed by professionals worldwide for quick and competent triage. The datasets used in these consultations, along with the insights generated among physicians, can then be used to further train the AI medical assistant.

Routine eye exams might soon involve a patient simply using a black box such as the kind used in the IRIS program and then receiving a multi-physician preliminary diagnosis in just hours.