Automated interpretation of retinal fundus photographs is an emerging tool that can help screen for diabetic retinopathy (DR), and new research suggests these grading systems are both feasible and a viable way to expand screening programs.
A recent prospective observational study was conducted at two eye care centers in India and included 3,049 patients with diabetes. The team compared automated DR grading system results with manual grading results, monitoring for the sensitivity and specificity of moderate or worse DR.
For moderate or worse DR, the study authors determined that the sensitivity and specificity for manual grading ranged from 73.4% to 89.8% and from 83.5% to 98.7%, respectively. The automated DR system’s performance, on the other hand, was equal to or greater than the manual grading, with a sensitivity ranging from 88.9% to 92.1% and a specificity ranging from 92.2% to 95.2%.
“While there are many avenues for future work, this study demonstrates the feasibility of using an automated DR grading system in health care systems and shows that the trained algorithm generalizes to this prospective population of Indian patients,” the study concludes.
|Gulshan V, Rajan RP, Widner K, et al. Performance of a deep-learning algorithm vs. manual grading for detecting diabetic retinopathy in India. JAMA Ophthalmol. June 13, 2019. [Epub ahead of print].