Machine learning can predict eyes at risk of developing diabetic retinopathy

An automated machine learning model could help identify eyes at risk for diabetic retinopathy (DR) progression based on ultra-widefield retinal images, a study published online on February 8 said. The research results were announced. JAMA Ophthalmology.

Paolo S. Silva, MD, PhD, of Harvard University in Boston and colleagues evaluated whether an automated machine learning model using ultra-wide-field retinal images predicts the progression of DR. The analysis included 1,179 anonymized ultra-wide-field retinal images with mild or moderate nonproliferative DR (NPDR) from 3 years of long-term follow-up.

The researchers found that the area under the model’s precision-recall curve was 0.717 for baseline mild NPDR and 0.863 for moderate NPDR. In the validation set, sensitivity was 0.72, specificity was 0.63, and accuracy was 64.3 percent for eyes with mild NPDR, whereas performance for eyes with moderate NPDR was 0.80, 0.72, and 73.8 percent, respectively.

In the validation set, 6 of 9 eyes (75%) had mild NPDR, and 35 of 41 eyes (85%) had moderate NPDR that progressed to 2 or more stages. In this model, all four eyes had mild NPDR that progressed within 6 months to 1 year, 8 of 9 (89%) had moderate NPDR that progressed within 6 months, and 20 eyes had moderate NPDR that progressed within 6 months. Seventeen of these (85%) were identified as having progressed within 1 year.

“The use of machine learning algorithms has the potential to adjust the risk of disease progression and identify patients at highest short-term risk, potentially reducing costs and improving vision-related outcomes.” write the authors.