People with diabetes have an increased risk of blindness. The research team compared the ability of two sets of computer programs to detect possible eye problems in 16,670 people with diabetes. Each of the two programs performed equally well.
The systems require a trained technician to use a digital camera to take pictures of the retina. The images are then transferred electronically to computers, which can automatically detect the small hemorrhages (internal bleeding) and signs of fluid that are hallmarks of diabetes damage.
"Our analysis shows that the computerized programs appear to be as accurate as a highly trained expert in determining if these initial signs of an eye problem are developing in someone with diabetes. Once the initial problems are found, an eye specialist can treat the patient," says researcher Michael Abràmoff.
Abràmoff said that among a group of 100 patients with diabetes, ten people would likely have diabetes-related eye problems. An ophthalmologist would have to check the eyes of all patients to find out who had problems. The computer programs, when given photos of the eyes of the same 100 patients, flag on average 20 people as possibly having diabetes-related eye problems. Thus, an ophthalmologist would need to see only the 20 people prescreened by the computer program instead of the original 100.
"The computerized programs allow ophthalmologists to spend time on patients who actually need care and provide better care to those patients," Abràmoff said. He noted the study had some limitations. For one, the images were prescreened to ensure the computers could analyze them. However, his research group has already developed the tools to automatically ensure adequate image quality before proceeding.
In addition, the number of people in the study who actually had diabetes-related eye problems was lower than what might be seen in other populations. Thus, Abràmoff said, it will be important to test the systems in larger groups. Last, the computer-based assessments were compared to assessments done by only one human reader at a time, which may not reflect a comparison to assessments by multiple readers.
MEDICA.de; Source: University of Iowa - Health Science