 This meta-analysis examines the overall diagnostic accuracy of machine learning, ML, in diagnosing diabetic retinopathy, DR, based on color fundus photographs, and determines that ML algorithms demonstrate high accuracy in detecting DR of various categories, with a pooled area under the receiver operating characteristic, OROC, ranging from 0.97 to 0.99. The study also found that neural network was the most widely used method, and had a pooled OROC of 0.98 for studies that used neural networks to diagnose more than mild DR. However, the results should be interpreted with caution due to methodology flaws in earlier published studies. This article was authored by Joe Su and Wu, T.Y. Alvin Liu, Wanting Su, and others.