IBM trains Watson to detect eye problems


It seems that hardly a week goes by nowadays without IBM Corp. finding a new application for its Watson artificial intelligence system.

The latest addition to the list is detecting eye problems, an area where the technology giant hopes to automate diagnosis in the same way that it’s doing over on the oncology front. The push is being led by a team from IBM’s Melbourne research center that revealed the first fruits of its effort today.

The team’s work focuses on harnessing Watson to identify early signs of glaucoma, which often goes undetected until the patient starts experiencing vision loss. Since the artificial intelligence had never before been used for diagnosing eye problems when the project launched in 2015, the team needed to start from the basics. The first step was feeding their system some 88,000 anonymized retina images from a clinical data exchange called EyePACS to familiarize it with the underlying anatomical structure.

Two years later, Watson has amassed a sufficient knowledge repository to distinguish between patients’ left and right eyes with a confidence level of up to 94 percent. IBM’s researchers also taught the artificial intelligence to overcome common image quality problems and avoid confusing a scanning error for a medical problem. Building on this foundation, the team then trained the platform to measure the ratio of the optic disk (the anatomical location of the eye’s “blind spot”) to the so-called optic cup, which is one of the main ways of assessing whether a patient suffers from glaucoma.

IBM claims that the platform can now make the comparison with “statistical performance as high as 95 percent.” Going forward, the company will work on further improving detection accuracy while broadening the scope of diagnosis to other common eye problems such as diabetic retinopathy and age-related macular degeneration.

Image: StockSnap