Skip to main content
. 2022 Feb 4;7(1):e000924. doi: 10.1136/bmjophth-2021-000924

Figure 3.

Figure 3

Examples of incorrect predictions with corresponding saliency maps (centre column) and GradCAM++ (right column) visualisation for each class. The heatmaps show examples of erroneous classifications and false positives. (A–C) The model assigns an image of diabetic retinopathy (DR) to the retinal vein occlusion (RVO) category. (D–F) The model erroneously assigned an image of RVO to the DR category. (G–I) The model mistakenly classified an image of DR to the healthy category. (J–L) The model classified an image from a healthy control as DR.