Keel S, et al. - Researchers undertook this cross-sectional study to systematically visualize the convolutional neural networks of two validated deep-learning models for the detection of referable diabetic retinopathy (DR) and glaucomatous optic neuropathy (GON). In 96 of 100 true-positive DR patients, the most important prognostic regions were lesions typically observed in cases of referable DR (exudate, hemorrhage, or vessel abnormality). Heat map visualization within traditional disease regions was noted in all 100 GON patients. Findings supported the validity of deep-learning models, confirming the reliability of a visualization method that may promote clinical adoption of these models.