Diabetic Retinopathy Detectors Equally Effective

New algorithmic detector comparable to detector used in large-scale early-detection project

THURSDAY, May 20 (HealthDay News) -- A new automated detection algorithm appears to be as effective at detecting diabetic retinopathy as an established algorithm used in a large early-detection project, according to a study published online April 16 in Ophthalmology.

Michael D. Abramoff, M.D., Ph.D., of the University of Iowa Hospitals and Clinics in Iowa City, and colleagues performed a study in which a single retinal expert analyzed two fundus images from each eye of 16,670 individuals with diabetes who were not previously diagnosed with diabetic retinopathy. The outcomes of the two algorithmic detectors were applied to the data and compared. One algorithm won the 2009 Retinopathy Online Challenge Competition (Challenge2009), and the other is used in EyeCheck, a large computer-aided early diabetic retinopathy detection project.

Examination results indicated that more than minimal diabetic retinopathy was detected with both algorithms, with an area under the curve (AUC) of 0.839 for the EyeCheck algorithm and an AUC of 0.821 for the Challenge2009 algorithm, a statistically nonsignificant difference. The researchers found that the AUC for detection was 0.86 -- equal to the theoretically expected maximum -- if either algorithm detected diabetic retinopathy in combination. The specificity of the EyeCheck algorithm was 47.7 percent and the specificity of the Challenge2009 algorithm was 43.6 percent, at a sensitivity 90 percent.

"Diabetic retinopathy detection algorithms seem to be maturing, and further improvements in detection performance cannot be differentiated from best clinical practices, because the performance of competitive algorithm development now has reached the human intrareader variability limit," the authors write.

Two authors disclosed serving as owners of EyeCheck.

Abstract
Full Text (subscription or payment may be required)

Related Stories

No stories found.
logo
www.healthday.com