With diabetes on the rise, current efforts are in-person screening and grading of diabetic retinopathy usually fails to keep up with the demand. Now, research suggests artificial intelligence (AI) may be able to help triage patients to assist with the screening efforts. In a new study, researchers evaluated the performance of an automated AI algorithm for triaging retinal images for signs of disease compared with human grading that followed a standard national protocol for reference.

The researchers used both manual and automated grading with machine-learning software to assess retinal images from 30,405 consecutive screening episodes. They measured screening performance (sensitivity, specificity) and diagnostic accuracy, using human grades as the reference standard.

The AI algorithm had 95.7% sensitivity for referable retinopathy. Within those images, the AI had 98.3% sensitivity for mild-to-moderate non-proliferative retinopathy with referable maculopathy, 100% sensitivity for moderate-to-severe non-proliferative retinopathy and 100% sensitivity for proliferative disease. The algorithm agreed with the human grading of no retinopathy in 68% with a specificity of 54% when combined with non-referable retinopathy.

The authors concluded that the system demonstrated safe levels of sensitivity for high-risk retinopathy in a real-world screening service. The AI’s specificity was comparable to that of human graders. “AI machine learning and deep learning algorithms such as this can provide clinically equivalent, rapid detection of retinopathy, particularly in settings where a trained workforce is unavailable or where large-scale and rapid results are needed,” the researchers note.