FDA approves first AI tool for detecting retinopathy, NIH shows machine learning success in imaging

The U.S. Food and Drug Administration and National Institutes of Health each made moves in the artificial intelligence and machine learning realm this week that potentially foretell what hospitals should expect in the not-so-distant future. 

FDA has given its blessing to the marketing of the first medical device that uses artificial intelligence to detect more than just a mild level of diabetic retinopathy, an eye disease found in adults with diabetes. The move has implications for enhanced patient access, which in turn has the potential to both increase collections and improve care management among the diabetes population.

[Also: Where AI has the most promise for reducing healthcare costs]

Diabetic retinopathy occurs when high levels of blood sugar lead to damage in the blood vessels of the retina, the light-sensitive tissue in the back of the eye. It's the most common cause of vision loss among diabetic American adults, and the leading cause of vision impairment and blindness among working-age adults.

Early detection is key, so the Food and Drug Administration's decision to green-light the technology -- which can be used in a primary care doctor's office -- has the potential to improve population health efforts.

The device, called IDx-DR, is a software program that uses an artificial intelligence algorithm to analyze images of the eye taken with a retinal camera called the Topcon NW400. It's the first device authorized for marketing that provides a screening decision without the need for a clinician to also interpret the image or results, which makes it usable by healthcare providers who may not normally be involved in eye care.

The technology was able to correctly identify the presence of more than mild diabetic retinopathy 87.4 percent of the time, and was able to correctly identify those patients who did not have more than mild diabetic retinopathy 89.5 percent of the time.

The IDx-DR isn't the only new technology to harness the power of artificial intelligence. 

Researchers from Massachusetts General Hospital, Martinos Center for Biomedical Imaging, and Harvard University recently set out to improve MRI image reconstruction by harnessing the power of machine learning.

The researchers used recent advances in technology, such as more powerful graphical processing units in computers and artificial neural networks, to develop an automated reconstruction process. They named it Automap, for automated transform by manifold approximation. To train the neural network, the team used a set of 50,000 MRI brain scans from the NIH-supported Human Connectome Project.

The team then tested how well Automap could reconstruct data using a clinical, real-world MRI machine and a healthy volunteer. They found it enabled better images with less noise than the conventional MRI, and the signal-to-noise ratio was better than in conventional reconstruction. Automap also performed better on a statistical measure of error known as root-mean-squared-error (6.7 percent versus 10.8 percent), and was faster than the manual tweaking now done by MRI experts.

Researchers said the new method could be applied to improving the quality and speed of various imaging methods, both medical and nonmedical -- which could ultimately help hospitals reduce costs.

Twitter: @JELagasse
Email the writer: jeff.lagasse@himssmedia.com