Reported accuracies were 86% (Random Forest) and 96% (convolutional neural networks), positioning retinal imaging as a candidate scalable tool for underserved populations. AI-powered polarized-light ...